The paper concentrates on psycholinguistic processes which occur while decoding speech from the acoustic signal to a complete recognition of the word. The acoustic signal reaching the hearer fails to reveal any clear-cut phonemic boundaries or invariability, therefore different perception models refer to different sources in speech categorization. The reason for the fact that phonemic categories are so strongly blurred in the signal is coarticulation, which, despite its disruptive effect on the structure of the signal, appears to be crucial in increasing the effectiveness of speech recognition. Having processed the signal into distinct speech categories, the hearer searches for an appropriate lexeme in their lexicon. The process appears to rely strongly on two aspects; competition and neighbourhood. Lexemes congruent with the incoming speech signal are activated in parallel and compete for recognition. Lexemes in dense neighbourhood are activated differently from lexemes in sparse neighbourhood. In its final parts, the article discusses how the ability to write and read influences the phonological representation of words.