Used in
cryptanalysis to identify possible values for the letters/icon/pictures in the
encrypted text when a
substitution cipher is used. The idea is that the most common character is probably one of the most common
letters in the suspected
plaintext language.
Take this a step forward and you can analyze how letters are related to each other (what letters can be found next to it in the language and how often they are found together). Using this information you can make better guesses at which characters represent which letters by how sociable they are to other characters. For example if a character is found next to many different ones it is probably a vowel because vowels are very sociable letters. If the character is found next to a limited number of vowels or in limited combinations then it is probably a consonant.
Take it another step and you can analyze sequences of letters and how often they are found together.
Basically an analysis of the plaintext language must be performed and statistical information gathered to apply educated guesses based on the frequency of letters and letter groupings.
One last thing: frequency anlysis was discovered by the Arab scientist Abu Yusuf Ya`qub ibn Is-haq ibn as-Sabban ibn `omran ibn Ismail al-Kindi (or Al-Kindi for short) in the 9th century.