Section I
Generation and Use of Mixed Cipher

Mixed Cipher Alphabets
Mixed cipher alphabets differ from standard alphabets in that one or both sequences
are mixed sequences. A mixed sequence is any sequence not in normal alphabetical
order. The two main types of mixed sequences are systematically mixed and random
mixed sequences.

a. Systematically mixed sequences are produced by an orderly process based on easily
remembered keywords, phrases, or simple rules. There are a number of mixed
sequence types, which will be explained in this section. Their advantage is that the
keys can be easily memorized and reconstructed for use when needed. Their disad-
vantage is that the orderliness in construction can be used by the opposing
cryptanalyst to aid in their recovery.

b. Random mixed sequences are not based on any orderly generation process. They
can be produced by various means ranging from pulling the 26 letters out of a hat to
complex machine generation. Their advantage is that their structure offers no help
to the opposing cryptanalyst. Their disadvantage is that the keys cannot be
memorized easily or produced from simple directions as systematically mixed
sequences can. They must be printed out in full and supplied to every user.

Keyword Mixed Sequences
One of the simplest types of systematic sequences is the keyword mixed sequence. The
sequence begins with the keyword, which may be a word or a phrase. Any letters
repeated in the keyword are used only once, dropping the repeating letters. After the
keyword, the rest of the letters are listed in alphabetic order, omitting those already


Repeated letters dropped: CRYPTOGAHI
Remaining letters added in normal order:



Repeated letters dropped: MILTARYNEGC
Remaining letters added in normal order:


Transposition Mixed Sequences
Transposition mixed sequences are produced by writing a letter sequence into a matrix
and extracting it from the matrix by a different route. The most common types are
called simple columnar, numerically keyed columnar, and route transposition
a. Simple columnar transposition is usually based on a keyword mixed sequence. The
keyword determines the width of the matrix that is used. The keyword is written as
the first row of a matrix and the rest of the sequence is written beneath it, taking as
many rows as necessary. The transposition mixed sequence is then produced by
extracting the columns of the matrix from left to right.

Keyword” ARTILLERY Keyword“ MORTAR

Keyword mixed sequence in matrix: Keyword mixed sequence in matrix:

Resulting sequence: Resulting sequence:

b. The numerically keyed columnar transposition mixed sequence differs from the
simple columnar only in the way it is extracted from the matrix. Instead of
extracting the columns left to right, the order of the columns is determined by a
numerical key based on the keyword. After constructing the matrix, the letters in
the keyword are numbered alphabetically. The columns are then extracted
according to the resulting numerical key.


Resulting sequence: Resulting sequence:

c. Route transposition sequences are formed by any other systematic way of entering
sequences into a matrix and extracting them from a matrix. They can be based on
standard or keyword mixed sequences. The samples in Figure 4-1 show some of the
common routes that can be used. The last two omit the letter J for the convenience
of a square matrix.

Decimation Mixed Sequences
Decimation mixed sequences are produced from a standard or keyword mixed
sequence by counting off letters at a regular interval.

a. As an example, consider decimating a standard sequence at an interval of 3. The
new sequence begins with the first letter of the basic sequence, in this case, A. The
second letter of the new sequence is the third letter that follows from the basic
sequence, D. Every third letter is selected until the end of the basic sequence is

Basic sequence:

Resulting decimated sequence:

The count then continues as if the sequence were written in a circle. The next letter
after Y, skipping Z and A, is B. The complete resulting sequence is shown below.

b. The interval should have no common factors with the length of the sequence. Since
any even number has a common factor of 2 with 26, only odd numbers are selected
with 26 letter sequences. Intervals with common factors are not selected, because
the count will return to the starting point again before all the letters are used. The
interval should also be less than half the length of the sequence, because larger
numbers will just duplicate in reverse order the sequence produced by a smaller
number. An interval of 23, for example would produce the same sequence as an
interval of 3, but in the reverse order. For a 26 letter sequence, the only usable inter-
vals are 3, 5, 7, 9, and 11. By counting either left to right or right to left, all the basic
decimated sequences can be produced.

c. Study of this method of decimation is particularly significant, because the solution
of some types of polyalphabetic ciphers can yield sequences in a decimated order
instead of the original order.

d. An alternate method of decimation is occasionally encountered. In the alternate
method, each letter is crossed off as it is selected and that letter is not counted
again. The restrictions on intervals do not apply to this method, because the
starting letter can never be reached again. This method is used less, because it is
subject to mistakes in the counting process that are hard to detect and correct.

Types of Mixed Cipher Alphabets
As mentioned at the beginning of this section, a mixed alphabet is any alphabet that
uses one or more mixed sequences. The simplest types are those which use a standard
sequence in one component and a mixed sequence in the other. These are the easiest
for a cryptanalyst to reconstruct. Next in order of difficulty are those in which the
same mixed sequence is used in the plain and cipher components. Most difficult are
those in which two different mixed sequences are used. The next section shows how to
recover each of these types of alphabets.

Section II
Recovery of Mixed Cipher Alphabets

Alphabet and Plaintext Recovery
Although this manual separates the techniques of alphabet recovery from plaintext
recovery, the two processes will usually occur simultaneously, each supporting the
other. When an orderly structure is found in an alphabet as individual letters are
recovered, the orderly structure often helps make more plaintext recoveries. The
techniques explained in this section will be used in the next section.

a. You usually begin reconstruction by recording recoveries in the form of an enci-
phering alphabet. An enciphering alphabet is one in which the plaintext component
is arranged in A through Z order. Ciphertext letters are written in the cipher compo-
nent paired with their plaintext equivalents in the plain component. The plaintext
can be either the top or bottom letters, but whichever you select, you should follow
it consistently in the alphabet as well as the cryptogram. Inconsistency leads to
errors. In this manual, plaintext is placed above ciphertext.

b. A deciphering alphabet is one in which the ciphertext is written in A through Z
order. Rearranging the alphabet into deciphering order is sometimes helpful in
alphabet recovery.
c. Whenever systematically mixed alphabets are used, you should attempt to recover
the systems and keys in use. The same sequences are often reused, either at dif-
ferent alignments of the same alphabet or in combination with other sequences. The
solution can be reached much quicker when you recognize and take advantage of
previous recoveries.

Reconstruction of Alphabets With One
Standard Sequence
Whenever one of the two sequences is a standard sequence, recovery of the system used
to produce the other sequence is made much easier.

a. The easiest type to recognize is the keyword mixed sequence. Any keyword mixed
sequence has two parts”the keyword and the alphabetic progression. If you find
that recovered letters are falling in alphabetic progression consistently in a portion
of the sequence, it is probably a keyword mixed sequence. In this case, you can
narrow down the possibilities of unrecovered letters. Consider the following par-
tially recovered alphabet.

(1) The letters DFGI appear to be part of the alphabet section of the cipher
sequence. The alphabetic progression continues at the left with the letters S
and Z. All the other recovered letters appear to be part of the keyword. Between
the H and the D there is room for only two of the letters at the beginning of the
alphabet”A, B, and C. At least one of these must be in the keyword, leaving
the other two as probable equivalents of plaintext P and Q. Similarly, there is
space for only three letters between S and Z. T and V already appear, so the
spaces must be filled by three of the four letters, U, W, X, and Y. Given these
limitations, recovery of more plaintext is likely. Continuing the example, con-
sider that plaintext C, F, L, P, W, and Y are recovered next.

(2) These recoveries enable several more probable letters to be placed by
alphabetical progression.

(3) At this point, we can see that A and E must be in the keyword, because there is
no room for them in the alphabetic progression. U or W must be in the keyword,
because there is only room for one of them between S and X, and V is already
placed. Similarly, M or N and Q or R must be in the keyword. Q is unlikely,
even though U is available to pair with it. Placing Q and U anywhere in the
blanks in the keyword suggests nothing further. R must be in the keyword, then.

(4) The letter after L in the keyword must certainly be a vowel or the keyword
would be unpronounceable, and that vowel represents plaintext G. With the
possibilities narrowed down this far, you might be able to spot the keyword

without referring back to the cryptogram that produced the partially recovered
alphabet. The complete alphabet looks like this.

b. Recovery of decimated sequences is a straightforward process of trying out inter-
vals. Just as a decimated sequence is produced by counting at a regular interval,
the original sequence can be recovered by counting at a regular interval, too. A par-
tially recovered alphabet with a suspected decimated sequence in the cipher com-
ponent could look like this example.

(1) To determine if this is a decimated sequence, various intervals can be tried.
The recovered letters suggest one obvious possibility. The letters V, W, and X
all appear among the recovered letters. If they were in order in the base
sequence used to generate the decimated sequence, they should reveal the in-
terval. The interval from V to W and from W to X is -5 in each case. A trial
decimation at -5, beginning with V produces the following sequence.

(2) This sequence of letters appears to be a keyword mixed sequence. The keyword
appears after the VWX and alphabetic progression resumes with the F and the
KM. Once you recognize this structure, you can use it to assist in further plain-
text recoveries just as in the first example shown in paragraph 4-7a. The original
basic sequence used to produce the decimated sequence is shown below.

c. Simple transposition mixed sequences often resemble decimated sequences. You
will often see a regular spacing of adjacent low frequency letters, just as we saw
VWX in the previous example. This is not caused by a decimation interval, but by
the regular length of columns separating the letters. Recovery of the generation
method of transposition mixed sequences is accomplished by rebuilding the original

The almost regular spacing of the letters V, X, Y, and Z resembles a decimated
sequence, but the interval is not constant. This almost, but not quite, regular
spacing is an indication of simple columnar transposition. The letters V, X, Y, and
Z are probably the bottom letters in their columns of the original matrix. W, which
has not been recovered, probably occurs in the keyword, because there does not ap-
pear to be room for a column ending with W. Analysis of this type of sequence
proceeds by rebuilding the columns. Placing the letters V, X, Y, and Z in sequence
with their preceding letters as their columns, produces this partial result.

Now the initial reconstruction appears successful. The rows above VXYZ also show
alphabetic progression developing. Q can be inserted in the next to last row with
confidence. The next step is to place the rest of the letters into columns that would
continue the structure in a logical way. A little trial and error will show that the
columns before the V column end with T and U. The U was not the top of the V
column, but the bottom of the preceding column.

The longer columns belong on the left. Shifting these columns produces this result.

The matrix is now in its original form. L, M, and N can be placed between K and O.
Either H or J can be inserted between F and K and the remaining letter belongs in
the keyword in the top row. S and W are in the keyword, because they are missing
from the alphabetical progression. That leaves A, B, or C for the remaining letter of
the keyword, with the other two on the second row. Since only one vowel has been
found in the keyword up until now, A probably belongs in the keyword with B and C
filling the blanks in the second row. Trial placements of A, S, and W together in the
first row blanks, together with either H or J in the remaining space leads to the con-
clusion of JIGSAW as the keyword.

d. The recovery of numerically keyed columnar transposition sequences is the same as
for simple columnar transposition, except the columns are not in order in the
sequence. The next example shows the recovery of this kind of transposition mixed

This problem is again best approached through the end of alphabet letters. V, W,
X, Y, and Z have all been recovered, and they make a good starting point. V, W, X,
Y, and Z are placed in a row with their preceding letters above them in columns.

This time no alphabetic progression appears, even if we consider that one or two of
the columns might be misplaced. In this case, the next thing to consider is that the
sequence may be reversed. Selecting the letters to the right of V, W, X, Y, and Z
instead of the left produces the following example.

This setup is clearly correct. Next, we add the two short remaining segments.

Moving the short columns to the right and filling in the missing letters produces the
following matrix.

The final step is to recover the numerical key. If normal methods are used, it should
be produced by the keyword and should show the actual order in which the columns
were extracted. Numbering the letters in the keyword in alphabetical order and
comparing them with the cipher sequence in the alphabet confirms that this
method was used. Since the sequence was reversed, the order of columns in the
cipher sequence appears in right to left order beginning with the cipher letter B.

e. One type of transposition sequence remains to be considered. When a route
transposition process is used, the solution is to try to reconstruct the original routes.
In examining attempts to solve the matrix by rebuilding columns, be alert to entry
routes other than by rows. Look for spirals, diagonals, and alternate horizontals or
verticals. If rebuilding the columns produces no results, consider rebuilding spiral,
diagonal, or alternate row or column routes. This manual does not show examples of
these approaches, but if you encounter this situation, approach it logically and try
various approaches until one succeeds. The techniques of solving route transposi-
tion ciphers explained later in this manual will help in this process.

f. Each of the preceding examples was approached as if we knew, perhaps from past
history, what types of sequences were used. We assumed that the plain component
was a standard sequence, and the cipher sequence could then be readily reconstruc-
ted by itself. It is common, in approaching a cryptanalytic problem, to assume the
simplest case and only to move on to more complex possibilities when the simplest
case must be rejected. A great deal of time can be wasted by assuming something is
more complicated than it is.

g. The next simplest case is where the cipher sequence is a standard sequence and the
plain sequence is mixed. When reconstruction attempts fail because you started
with an enciphering alphabet, rearranging the alphabet into a deciphering alphabet
may yield results. Once rearranged, the solution is approached just as we did in the
above examples. Look for short alphabet progression to indicate keyword mixed
sequences. If that is not found, see if a decimation was used. If decimation was not
used, try reconstructing the columns of a columnar transposition. Remember to try
forward and reversed sequences.

h. If none of these approaches yields results, either with an enciphering alphabet or a
deciphering alphabet, other approaches are called for. Either there are two mixed
sequences, a more complex process was used, or random sequences were used.

Reconstruction of Alphabets With Two Mixed
Recovering alphabet structure when both sequences are mixed is more difficult than
the previous examples. You are much less apt to be successful with only partial
recoveries. Where the alphabet could be reconstructed during the solution of the plain-
text in the previous examples, reconstruction of an alphabet with two mixed sequences
must usually wait for the full solution of the plaintext. The examples in this section
will begin with a fully recovered, but not reconstructed, alphabet.
a. The easiest type to recover with two mixed sequences occurs when both sequences
are keyword mixed, as in the next example.

Enciphering and deciphering forms of the same alphabet are shown. The under-
lined portions show substantial alphabetic progression in both, which is typical of
alphabets with keyword mixed sequences. A transposition or decimation would not

produce such an obvious progression. The underlined portions in both alphabets are
probably in their original form. The remaining plain-cipher pairs are out of order.
Your task is to reconstruct the original order. The usual approach at this point is to
try to extend the alphabetic progression outward from the obvious progression. In
this case, the enciphering alphabet shows two long alphabetic strings of cipher let-
ters, HIJKLMNO and WXYZ, which must have some or all of the letters
PQRSTUV in between. Similarly, the deciphering alphabet shows plain-
text strings ABCD and STUVWXYZ, and some or all of the letters
EFGHIJKLMNOPQR must be in between. Suppose the cipher letters PQRSTUV
belong in exactly that order. If that is the case, then the plaintext letters
GOMPHER must also be in the right order, preceding ABCD. We expect to find the
keyword immediately before the beginning of the alphabetic sequence.
GOMPHER, while not a recognizable word may be close to it. If we try GOMPHER
as a keyword, then the remaining letters must be in alphabetical order. Adjusting
the alphabet so GOMPHER is a trial keyword will produce this arrangement.

Now the cipher sequence shows a recognizable word, BADGE, but the solution is
incomplete. If we move the M-R pair so that plaintext M fits in alphabetic order
instead of the keyword, we see the following alphabet.

This rearrangement is the original sequence of the alphabet.
b. When transposed or decimated sequences are used in the alphabet, the solution is
much more difficult. The alphabetic progression used in the previous example is
not available to assist with reconstruction. A solution is still possible in many cases,
however. When both sequences are the same sequence in the same direction, the
alphabet can often be recovered quite readily.

(1) Reconstruction begins with a process called chaining. Use the plain-cipher
pairs to create a 26 letter chain by linking the cipher letter of each pair to the
pair with the same plaintext letter. Any pair can be used as the starting point.
Beginning with the plaintext A-ciphertext L pair (abbreviated Ap-Lc) next find
plaintext L. Plaintext L equals ciphertext W (Lp-Wc), producing a partial

chain of ALW. Continuing with Wp-Bc, the chain is extended to ALWB. Con-
tinue adding links to the chain until you return to the original letter A. The
complete chain is shown below.

(2) Since we were able to produce a 26 letter chain, there is a strong indication that
the same sequence was used in both components. With different sequences, the
chances of producing such a chain are very low. Unrelated sequences will almost
always return to the starting point before using all 26 letters. The alphabet in
paragraph 4-8a, for example, produces separate 23 and 3 letter chains.
(3) The sequence produced by chaining an alphabet with two identical sequences in
the same direction will always either be the original sequence or a decimation of
the original sequence. This narrows the possibilities for the original sequence
down to six. The chained sequence and its five possible decimations are listed

(4) If the original sequence was a decimated sequence, the basic keyword or stan-
dard sequence used to generate the decimated sequence would be one of the
above. Since none of them are either standard or keyword mixed, the original
sequence was probably transposed. Approaching each sequence above with
transposition in mind, the letters V, W, X, Y, and Z have been underlined in
each, searching for a basis to rebuild the columns. The last sequence (decima-
tion 11) yields the following matrix.

(5) When the same sequence is used in the same direction in both components of
the alphabet, a 26 letter chain will only be produced half of the time. When the
two sequences are staggered by an odd number of letters, a 26 letter chain
results. When the two sequences are staggered by an even number of letters, two
separate 13 letter chains result. These can sometimes be recovered, too, but the
solution is more difficult.

c. The chaining technique can also be used with alphabets with different sequences in
the two components if they are reused at different alignments. Consider the next
two alphabets, recovered at different times on the same day.

(1) To test if the same alphabet was used, chain the cipher sequences against each
other. In the example, chain A of the first to T of the second, T of the first to N
of the second, and so on. This produces the following chain.

(2) This confirms that the two alphabets used the same sequences at different
alignments. If chaining produced anything but one 26 letter sequence or two 13
letter sequences, they are not the same alphabet.

(3) Write all possible decimations, as before.

(4) The decimation of 7 produces a sequence that almost looks as if it were the
original. This can happen when the decimation interval and the column length
of a transposed sequence are the same except for one long column. The correct
sequence is a decimation of 9 read in reverse.

The sequence used to generate the simply transposed sequence was a keyword
mixed sequence based on LEMON.

(5) The plaintext component can be reconstructed now that the correct ciphertext
sequence is known. We start with the decimated sequence. Since the sequence
with a decimation of 9 was used in reverse to recover the keyword LEMON, we
will list it in reverse.

Either of the two alphabets given at the start of this problem can be used to
reconstruct the plaintext sequence. The first alphabet is repeated for reference.

We now rearrange this alphabet so that the cipher sequence is in the same order
as the recovered decimated sequence.

d. The chaining techniques introduced in this section are also used in the solution of
polyalphabetic ciphers. They will be further developed in Part Four.

Section III
Solution of Monoalphabetic Unilateral
Ciphers Using Mixed Cipher Alphabets

Preparation for Analysis

The first step in approaching the unsolved cryptogram is to prepare a worksheet.
a. If prepared by hand, one-fourth inch or one-fifth inch cross section paper (graph
paper) should be used if possible. Hand lettering should be clearly printed in ink.
The cryptogram should be triple spaced vertically to leave room for writing. If a
copying machine is available and local security rules permit, the worksheet should
be copied after preparation to permit a restart with a clean worksheet whenever

b. Generally, you will want to prepare at least a unilateral frequency count. Other
special frequency counts may be needed also, as will be explained later. If you are
unsure of system identification, you may want to calculate the φ IC. Computer sup-
port, if available, can save a lot of time at this step.

c. Next, you should scan the text searching for repeated segments of ciphertext.
Underline all repeats you find of at least three letters in length. You may find it
useful to underline two letter repeats, too.

d. If you have more than one cryptogram that appears to have been enciphered with
the identical system, prepare a worksheet for each. Compare peaks and troughs of
frequency counts to see if they are similar. If so, look for repeats between messages
as well as within messages. Repeats between messages are another indication that
the identical system was used. The more repeats you find, the easier the solution
will be.

e. If you are still in doubt whether two cryptograms have been enciphered by the same
system, there is a simple statistical test available, similar to the phi test. The chi
test or cross product test compares two frequency distributions to determine the
probability that they are from the same alphabet. The frequency of each letter in
one distribution is multiplied by the frequency of the same letter in the other dis-
tribution. The results of all the multiplications are added to produce the chi value.
Chi is the Greek letter that looks like an X. The formula for the chi value is”

The expectation with a random match is l/26th of the product of the total letters of
each, or”

With a correct match, the expected value is .0667 times the products of the total
letters, or”

The results can also be expressed as an index of coincidence, the usual form if
produced by computer support. The formula for the cross IC, as it is called is”

With a correct match, the expected IC value, as with the phi text is 1.73. If you
match two alphabets and the X IC is close to 1.73, the chances are that they were
enciphered with the same alphabet. Figure 4-2 illustrates a completed chi test.

f. As with any statistical test, you should use this as a guide only, and take all other
available information into consideration, too, For example, if you find several long
repeated segments of text between two cryptograms, it is probably a waste of time to
calculate a chi test by hand. You already have the evidence you need to make a deci-
sion as to what approach you will use to reach a solution.

Approaches to the Solution
There are two basic approaches to the solution”the probable word method and the
brute force approach. The probable word method is to try to gain a quick entry into the
system by correctly assuming a portion of the plaintext. The brute force approach is to
systematically narrow down the possible keys to the system and then force a solution
by exhaustively trying all those possible keys. The method in the previous chapter of
solving standard alphabet systems through trying all possible decipherment is a good
example of the brute force approach. In practice, the solution of any given system is
likely to use a combination of the two approaches.

Solution With Known Sequences - Completing
the Plain Component Sequence
When the sequences used in an alphabet are known, a quick forced solution is possible.

a. Although mixed alphabets are used instead of standard ones, the solution is exactly
the same as that explained in paragraph 3-7b.
(1) Set up the known alphabet at any alignment.
(2) Perform a trial decipherment (pseudotext).
(3) Using the trial decipherment as the letters at the head of the columns, generate
all possible decipherment by listing the plain component sequence vertically
for each column.

b. Figure 4-3 illustrates the solution of a cryptogram with known sequences using the
above steps.

Probable Word Method
The probable word method of solution depends on your being able to correctly identify
a portion of the plaintext. When you can do this, you can begin to reconstruct the keys.
The partial key recoveries lead to more plaintext recoveries, and by working back and
forth between keys and plaintext, you can complete the solution. There are many ways
in which you can identify plaintext. The more you know about the senders of
enciphered traffic and the situation in which it was sent, the more likely you are to be
able to assume plaintext correctly.

a. Stereotypes. Military organizations tend to do things in standard ways. Rules for
message formats are likely to be used. Standard forms are likely to be used for
recurring needs. When you learn enough about the sender™s standard ways of doing
things, you can use those standards. Standard formats are most likely to be found
in message beginnings and endings. Messages are likely to begin with addressees,
message subjects, security classifications, and references to other messages.
Messages are likely to end with signatures or unit identifications. These stereotypes
are bad security practices, but difficult to avoid.

(1) Consider the following example of a message where stereotypes can be used to
achieve a quick solution. The previous message from the same sender, already
recovered, began, TWO PART MESSAGE PART ONE. The text gave the
itinerary of a visiting team of officers from an allied country, but was incom-
plete. A mixed alphabet was used with the previous message, but it has
changed with the new message.

(2) The first and last groups (ZZZZZ) are obviously not part of the text of the
message. They are probably indicators of some kind.

(3) We begin by preparing the following worksheet with a frequency count and
underlined repeats. The indicator groups are not included in the frequency

(4) If this is a follow-on to the message that began, TWO PART MESSAGE PART
ONE, we would assume that it would begin TWO PART MESSAGE PART
TWO. The underlined repeats are positioned perfectly for the repeated words
TWO and PART, so the assumption seems well borne out.

(5) Next, we enter the assumed text in the message and the alphabet. Using those
recovered values throughout the message produces the text shown below.

(6) From the recovered ciphertext letters, it appears that the cipher sequence is
keyword mixed. On that basis, ciphertext G and J are placed in alphabetical

(7) Several possibilities for additional plaintext appear in the message with these
additions. You may see other possibilities but for illustration, we will add the
letters for the word COMMANDING appearing at the end of the message.

(8) Additional placements are possible. Ciphertext Y belongs between X and Z. P
and Q fit between N and S. U, V, and W fit between Sand X. The first word on
the second line appears to be WILL. The phrase SIXTEEN THIRTY HOURS

Only the ciphertext letters A, B, and C remain to be placed. Of those, only A is
used in the text, and it appears to be part of the commander™s name. If C is
placed as part of the keyword ROCKET and A and B placed in alphabetical
order, the commander™s name becomes R L JONES. The plaintext is TWO
plete alphabet is shown below.

b. Exploitation of Numbers. Not all cryptograms will include such stereotyped
beginnings and endings. Without these stereotypes, repeated words in the text offer
another possible point of entry. Spelled out numbers are often easy to recognize
when they repeat in messages, as shown in the next example.

(1) The pattern of consecutive short three- to five-letter repeats is characteristic of
numbers. Numbers tend to occur with each other in such things as grid coor-
dinates, times, and quantities. In the above example, the repeated RSZNN
must be THREE, the only five letter number to end in a double letter. We begin
by placing THREE in the alphabet and entering other occurrences of the same

(2) The recovered letters suggest additional numbers. RCW, which begins with
plaintext T must be TWO. GNZW, which includes ER as the middle two letters
must be ZERO. EUD, which has no letters in common with THREE, TWO, or
ZERO, can only be SIX.

(3) Several more possibilities can be placed at this point. Ciphertext F can be
placed between D and G in the cipher sequence as the alphabetical structure
begins to appear. The last word of the message is apparently HOURS, needing
only the U to complete it. The partially repeated FOUR can be seen at the end
of line two, and SEVEN follows TWO on the third line.

(4) The first word is MOVE. Q can be placed between P and S in the cipher
sequence. The word BY completes the third line. With ciphertext K placed from
the word BY, ciphertext L and M can also be placed.

(5) COORDINATES online one provides the plaintext letter A as ciphertext J.
With J placed in the alphabet, the letter I must be in the keyword, along with T,
which will not fit in the alphabetic progression. The keyword is therefore
HOWITZER. The complete plaintext is MOVE YOUR UNIT TO COOR-

c. Word Patterns. When neither stereotypical beginnings and endings nor repeated
numbers provide a point of entry, repeated words can often be recognized by their
patterns of repeated letters.
(1) Such words as ENEMY, ATTACK, and DIVISION have repeated letter pat-
terns that make them easy to recognize. They are even easier to recognize when
the words are repeated in the text. Underlining the repeats gives an indication
of where the words begin and end. For example, ATTACK and BATTALION
have the same pattern of repeated letters. If the ciphertext OGGORF is
repeated in the text, it is much more likely to be ATTACK than a portion of the
word BATTALION. It could also be EFFECT, ATTAIN, or a number of other
(2) In the case where two or more words have identical patterns, such as ATTACK
and EFFECT, letter frequencies can help to decide between the possibilities. If
the letters O and F of OGGORF are high frequency letters and the rest are fairly
low, it is more likely to be EFFECT than ATTACK. If all the letters are high in
frequency, ATTAIN is likely.

(3) Tables have been compiled of common pattern words for various languages to
assist in analysis. Table D-3 in Appendix D of this manual provides an English

language word pattern table. Word patterns are also called idiomorphs. There is
a formal procedure for recording word patterns, which is followed in the table.
When you find a pattern word repeated in a cryptogram, you can follow the
same procedure to record the pattern and then look it up in the table. The
procedure is this”
Find the first repeated letter in the pattern, and designate all occurrences of
that character with the letter A.

Continue lettering alphabetically from left to right, making sure that each
new character gets the next letter of the alphabet and each repeated charac-
ter gets the same letter.

Stop lettering when the last occurrence of the last repeated character is
reached. In the example, P is the last occurrence of the last repeated charac-
ter. The final character Z is not lettered.

Designate any characters before and after the pattern characters with dashes
to show the length of the word.

(4) To use the pattern, refer to Appendix D, Table D-3. The patterns are in
alphabetical order beginning on page D-19. The pattern ABCDEBFGAF is
located on page D-34. The only word listed for this pattern is H EADQUARTER
S. The extra letters at the beginning and end of the pattern, designated by the
dashes, fit HEADQUARTERS perfectly.

(5) The use of word patterns to solve a cryptogram is shown in the next example.

(6) The cryptogram shows all repeats longer than three letters. There are a number
of shorter repeats, too, which will be used if necessary. We begin the analysis by
deriving the word patterns for the longer repeats. The pattern and possible
words from Appendix D for each repeat are shown below.

(7) CROSSROADS is the only choice for the second patten. There is an extra let-
ter at the end of the repeat, but that may have been caused accidentally by a
repeated first letter of the next word in each case. Using CROSSROADS as a
trial starting point, we compare common letters with the other repeats. From
CROSSROADS, we see that cipher M equates to plaintext R, for example.
Examining the possible choices for the MQIATQVG repeat, only REGIMENT
is consistent with the Rp-Mc pair. Similarly, the Op-Uc and Dp-Wc pairs of
CROSSROADS are consistent with DIVISION for the WABANAUV repeat and
no others. The common plaintext N and I between REGIMENT and DIVISION
also equate to the same cipher letters (V and A) giving further evidence that we
are on the right track. Using the common letters between CROSSROADS,
REGIMENT, and DIVISION with the XGGXFS possibilities shows that either
ATTACH or ATTACK is consistent with the first three. We now place the
letters of CROSSROADS, REGIMENT, and DIVISION in the alphabet and

(8) With this start, you should be able to see many more possible plaintext words in
the text. TOMORROW, VICINITY, and ROAD JUNCTION all appear with

only one or two letters missing. Many spelled out numbers also appear. The
repeated NGUC is STOP, a common stereotype used in telegraphic text in
place of a period. EALL is WILL. XGGXFS must be ATTACK. The completed
plaintext is”
(9) Use of word patterns is a powerful tool to gain entry into a cryptogram. It will
not always work out as easily as the example shown here. Repeated letters do
not always represent repeated words. Many words that are used in messages will
not be found in the word pattern tables, particularly proper names. Be alert to
the patterns of repeated letters in names you would expect to find in message
traffic. If you can recognize the pattern of a word, it does not have to be in the
tables to use it.

Vowel-Consonant Relationships
When you can successfully discover plaintext words in a cryptogram, the solution
usually comes quickly. Sometimes you will encounter a cryptogram in which you can
find no basis to assume plaintext. You can find no stereotypes, no usable numbers, and
no repeated pattern words. In these cases, you can use the characteristics of the
language itself to determine individual letters.
a. Language Characteristics. Languages which use an alphabet to spell out words
phonetically produce exploitable letter relationships. To make words pro-
nounceable, vowels and consonants tend to alternate. We do not expect to find
many consonants or many vowels consecutively. In cases where they do, the
possibilities are limited to pronounceable combinations. Exploitation of these letter
relationships begins by determining which letters are consonants and which are
(1) Vowels tend to occur next to consonants. Consonants tend to occur next to
vowels. Each contacts the other more readily than it contacts its own type.
(2) Since there are more consonants than vowels in English, vowels tend to contact
more different letters than consonants do. A vowel will commonly contact a lot
of different consonants, whereas a consonant will tend to contact the smaller
number of vowels. By studying which letters contact each other and how many
different contacts each letter has, we can sort ciphertext letters into vowels and
consonants fairly reliably.

(3) To make use of these vowel-consonant relationships, we use a special kind of fre-
quency count which charts contacts as well as frequencies.

b. Trilateral Frequency Count. The trilateral frequency count is used to record, for
each letter in a cryptogram, the letter that precedes it and the letter that follows it.
Figure 4-4 shows a cryptogram and its trilateral frequency count. The pairs of letters
appearing in the column below each letter of the alphabet are the preceding and
following letters for each occurrence. For example, the YG that appears below the
letter A shows that the first A in the cryptogram occurred as part of the segment
YAG. Refer to the cryptogram itself, and you will see that the segment YAG occurs
in the second group of the message. Two numbers appear above each letter of the
alphabet. The top figure is the frequency of that letter, which is the same as the
number of pairs of letters in the column below it. The second number is the number
of different letters the basic letter contacts. This type of frequency distribution and
its supporting contact information take some time to prepare by hand, but they can
lead to the solution when other methods fail.

(1) The contact information is used to determine which ciphertext letters are
vowels and which are consonants. More often than not, the highest frequency
plaintext letter is a vowel, even when E is not the highest frequency letter. An
even more reliable indicator is the number of contacts. The letter that contacts
the most different letters will usually be a vowel. In the example in Figure 4-4,
ciphertext Y is likely to be a vowel for both reasons. The letters that Y contacts
most frequently are likely to be consonants.

(2) In cases where there are several letters all about the same frequency and no
letter stands out as a likely vowel, we can begin our approach through likely
consonants instead. All or most of the lowest frequency letters should be conso-
nants. The letters they contact most frequently are likely to be vowels.

(3) We can use either a likely vowel or the set of likely low frequency consonants as
our starting point. Whichever we start with, we will use both as the problem
develops. The object is to separate the consonants and vowels by plotting the
contacts of each in separate vowel and consonant line charts.

(4) For our example, we will pick the low frequency consonants as the starting
point. The process begins by charting the contacts of the lowest frequency
letters. We will begin with the letters that only occurred once in Figure 4-4“C,
H, J, L, O, and P. Draw a horizontal line two to three inches long and write the
selected letters above it. Draw a vertical line several inches from the center of
the horizontal line producing a T-shaped figure. This is the consonant line. The
contacts are charted on the line with the first letters of each pair to the left and
the second to the right. Each new contact letter is charted on a new row. With
the contacts for C, H, J, L, O, and P charted, the consonant line appears below.

(5) Continue adding the lowest frequency letters one frequency group at a time. We
first placed those with a frequency of one. Next add those with a frequency of
two. Continue with those with a frequency of three and so on. Stop when the
next frequency would represent more than 20 percent of the total. Going any
further raises the chance too high of including a vowel that would bias the chart.
If a vowel occurs only once or twice and is included, its influence will be small. If
it occurs five or six times and we include it, it could lead to wrong follow-on

decisions on vowels and consonants. In our example, there are 130 letters. We
want to keep our sample below 20 percent, or not more than 26 letters
altogether. On this basis, we can add the frequencies of 2, 3, and 4, but not 5.

(6) The consonant line now shows that the low frequency consonants contact the
ciphertext letter Y more than any other letter. The probability is very high that
this is a vowel. It is tempting to select the letter V as a vowel, but it is better to
proceed one letter at a time at this point.

(7) Using the letter Y and its contacts, we next begin construction of a vowel line. It
is charted exactly the same as the consonant line chart. The vowel line
including just the letter Y™s contacts is shown below.

(8) The vowel line shows us we were correct in not initially accepting the letter V as
a vowel. It contacts the low frequency consonants quite readily, but it also con-
tacts a vowel readily. It may be a consonant such as R, L, or N which easily

combines with other consonants. We will not try to place V in either line at this

(9) The letter W contacts Y six times and is a likely consonant. We will continue by
going back to the consonant line and adding W.

(10) The letter T now appears as a strong candidate for a vowel. It is second only to
Y in consonant contacts so far, and just as importantly, it does not contact the
already selected vowel at all. We add T and its contacts to the vowel line.

(11) The vowel line shows A and U as likely consonants. Adding these letters to the
consonant line produces the next diagram.

(12) B appears to be a vowel. This is reinforced by the letters BUUB in the first line
of the text. If U was correctly selected as a consonant, B is probably a vowel on
the basis of this letter pattern. It is a good idea at this point to return to the
text and underline all the recovered vowels.

(13) Examination of the vowel-consonant patterns in the text confirms additional
consonants. Double letters preceding or following the vowel are very unlikely to
be vowels. We can then assign ciphertext E and Gas consonants. The GGBG
segment on the first line could not all be vowels. EE occurs three times in the
text following a vowel.
(14) V appears to be a consonant from the number of contacts in the vowel line, and
its appearance between vowels in the segments YVB and TVY confirm it as a
consonant. Placing G, E, and V in the consonant line produces this diagram.

(15) The letters F, I, and S remain unidentified. At least one of these is likely to be a
vowel, since four of the letters are expected to be vowels and we have only iden-
tified three so far. Comparing the appearance of F, I, and S in the vowel and
consonant lines, we see that the letter I is the best candidate for a vowel. The
letter I does not appear on the vowel line at all, whereas, F and S directly con-
tact a number of the recovered vowels. We now underline I in the text and add
it to the vowel line.

(16) There are a number of directions you can take at this point. No single example
can demonstrate them all. Some of the approaches that can be tried are”
TO analyze vowel combinations to determine individual vowels.

To search for the plaintext consonants N and H. These two letters have
typical patterns of contact with consonants and vowels. N tends to follow
vowels and precede consonants. H tends to follow consonants and precede
vowels. In some cryptograms these features will be very evident in the vowel
and consonant line diagrams. In others, they will not stand out at all.
To recover double letters by frequency analysis. Plaintext LL is the most fre-
quent double consonant. EE and OO are the most frequent double vowels.
TO recover common word endings such as -ING and -TION, which often
appear as repeats even when complete words do not repeat.

(17) We will use several of these approaches to complete the solution of the sample
problem. First, one vowel combination appears in the cryptogram, the cipher-
text TB as part of the segment TGTBU. Referring to the two-letter frequency
data in Appendix A, page A-2, the most frequent vowel combinations are EE,
IO, OU, and EA. TB is not EE, because it is not a double letter. It is likely to
be one of the other three. IO is particularly significant, because it is usually
part of a -TION combination when it appears. The letters G and U, which
precede and follow BT in the text, are high frequency consonants and support
the -TION possibility. The letter T occurs again before G, which would
produce -ITION, a very good letter combination.

(18) If TGTBU is -ITION, the letter U may appear with the typical pattern of
plaintext N. Examining the occurrence of U in the vowel and consonant lines,
we see that U follows vowels more often than it precedes them. It also precedes
consonants more often than it follows. The differences are slight, but they help
to confirm the initial assumption.

(19) Ciphertext EE occurs three times. This is likely to be plaintext LL. Each time
it is preceded by ciphertext T, which we have tentatively identified as the
plaintext I. ILL is another good combination that appears as part of many
common words such as HILL and WILL.

(20) Y is the most common letter, and it is a vowel. While we would not usually
begin analysis by assuming the most common vowel is E, our tentative iden-
tification of I and O make this much more likely. If Yc is Ep, then the remain-
ing high frequency vowel, Ic, is probably Ap.

(21) Placing all the tentative recoveries in the cryptogram produces the next

(22) With the assumed letters filled in, two numbers stand out. ONE appears in the
second line, and NINE appears in the last line. Since numbers tend to occur
with each other, our next objective is to try to place additional numbers adja-
cent to these two. If we try SEVEN after ONE because of the -E-EN pattern, it
leads to the recovery of SIX before ONE and FIVE before NINE.

(23) All of the high frequency plaintext letters except R are now recovered. Vc is the
obvious candidate for Rp due to its high frequency and appearance in the text.
(24) Placing plaintext S, V, X, F, and R reveals this text.

(25) Many possibilities for plaintext appear now. ZERO, POSITION, RIVER
CROSSING, PREPARATORY, and FOUR can all be seen upon close

(26) Analysis of the cipher sequence shows it to be a simply transposed keyword
mixed sequence, which identifies Jp as Zc and Qp as Mc.