Among the most important questions in psycholinguistics today are the following: By which processes does man organize and understand speech? Which are the smallest linguistic units and rules stored in the memory and used in the production and perception of speech? Are the same mechanisms at work in both cases? Discussed in this paper are investigations and experiments in speech production and perception (linguistic performance), which provide data for the study of linguistic competence. In the works described, the speech-wave (and/or the articulatory gestures) have been studied and attempts have been made to derive, given the output, the underlying input (linguistic constructions), and the transformation involved. In the same way, studies of speech recognition have investigated the mechanisms involved when passing from acoustic signal (input) to linguistic constructions (output). In both cases, the presentation of empirically testifiable facts has been used to verity or refute linguistic hypotheses. It is suggested that a model of speech recognition in order to function satisfactorily must be able to (1) identify discrete elements in the continuous, unsegmented, multidimensional speech signal; (2) apply syntactic rules, and make predictions, corresponding to the use of context and similar aids; and (3) make a choice of the direction in which interpretation is to be sought. Language universals should be considered in the model of man as "sender and receiver." (AMM)