A method of coding a large file for information retrieval is discussed. Random superimposed coding of machine derived 'roots' of the full vocabulary is used to generate an easily updatable and very compact code file. No thesaurus or dictionary of terms is needed. High speed is made possible by the simplicity of the searching algorithm as well as the ability to make a search for several key words simultaneously. The simplicity of the search facilitates implementation of the system on a small computer with access to a large bulk storage device.