This paper describes a strategy for the development of a speech recognition system for Finnish that is based on several different knowledge sources which include auditory, phonetic, and linguistic details. Auditory knowledge is derived from the application of computational models which simulate the human peripheral hearing system. Phonetic knowledge is represented by rule-based analysis, parsing and classification of phonetically relevant units and structures from the output of the auditory model. Finally, linguistic knowledge is used to filter the word forms generated by the phonetic level by accepting or rejecting hypotheses through the use of morphological analysis. The main components of the system including the structure of the rules are presented.