The paper introduces the design and implementation of a text retrieval hardware unit. After surveying a number of hardware text retrieval packages, it proposes a linguistic approach, in which the retrieval is treated as a regular expression which can be recognised by a finite state machine. The design of a finite state machine interpreter is given, and the problems of compiling a regular expression for this interpreter discussed. The algorithm for the compilation is given. Performance figures for the hardware, when connected to real systems, are given.