Method for automatically abstracting a document in machine readable form consisting in storing in a dictionary memory (8) language terms commonly used in document preparation, comparing language terms from an input document received from an input register (16) with the stored language terms, selecting language terms from input document which do not compare, selecting language terms from input document which compare, coding the selecting language terms with the identity of the input document and storing the language terms in memory (12). When retrieving a document from storage, the processor (10) under the control of instruction memory (14) compares the words in an input query against the word index file in memory (12) and provides in register (18) the selected documents whose identification code corresponds to the highest retrieval value calculated using each identification code of each language term that compares.