Abstract: Recovering text from a damaged electronic file by scanning an arbitrary stream of bytes and extracting text that is encoded as ASCII or Unicode. A byte of the damaged file is read. The read byte may be interpreted using the ASCII encoding standard. The read byte and the immediately preceding read byte may also be interpreted using the Unicode character encoding standard. The interpreted byte(s) is classified based upon the likelihood that the byte(s) is actually text for the particular character set rather than a control character, damaged data, or an element other than a textual character. The classifications are used to adjust a likelihood counter for each character type. The likelihood counter may be an integer value that indicates the probability that a text run has been detected. A text run is a sequence of bytes that is believed to be undamaged text. Each likelihood counter is then examined to determine whether there is a text run for one of the character types.