The process of speech comprehension consists of two large phases: the perception of vocal phenomena corresponding to the system of linguistic signs, and the interpretation of that code system. Both phases are made up by several levels that collectively ensure the decoding of speech phenomena in a regular cooperation with one another. Speech comprehension is an active process whereby the hearer interprets the speech phenomena s/he has perceived at successively higher levels. For that process to work, certain mental representations must be shared by speaker and hearer. The notion 'mental representation' can be understood in various ways; it includes thoughts, ideas, wishes, percepts, conceptions, etc. Specifically 'linguistic' mental representations, on the other hand, are such that they contain linguistically relevant signs and functions of the individual's patterns of knowledge. Certain portions of linguistic mental representations may keep changing or being modified throughout the individual's lifetime. The present paper analyses, in several series of experiments, cases in which objective acoustic phonetic parameters seem to contradict the corresponding mental representations. The aim is to highlight the relationship between articulatory/acoustic differences and the invariant features that underlie them as well as that between mental representations and objective parameters. Four areas are investigated with respect to Hungarian: (a) the production, acoustic patterns, and perception of vowels; (b) the variants of the phoneme /h/; (c) the coarticulatory behaviour of (r); and (d) a phonological rule of coarticulation applying in spontaneous speech. On the basis of the results obtained, the author attempts to answer the question of how the (apparent) paradox of the contradiction between the objective physical reality of speech and its mental representation might be resolved.