During the last 20 years, several studies have demonstrated the influence of orthographic knowledge on speech processing. Nevertheless, we recently collected data suggesting that not all the phonological representations activated during recognition processes are sensitive to orthography. While an influence of orthographic representations is systematically observed in metaphonological tasks, in speech recognition this influence seems to depend on the presence of lexical process as well as on the time required by the recognition process. Using both French and Portuguese materials, we manipulated the orthographic consistency of words and pseudowords presented either in an auditory lexical decision task or in a shadowing task. The comparison of these two languages was of interest since they differ in their degree of orthographic (in)consistency, French orthography being much more inconsistent than Portuguese orthography. While inconsistent words produced longer auditory lexical decision latencies than consistent words, no word consistency effect was found in shadowing. Similar results were observed in French and Portuguese. As concerns pseudowords, an unreliable effect was found with the French material in both lexical decision and shadowing, but only in participants presenting the longer average response latencies. Further studies conducted in Portuguese also demonstrated that the consistency effect observed in lexical decision was due to the lexical rather than to the decisional component of the task. Indeed, the comparison of two situations in which the shadowing response was made contingent upon either a lexical or a phonemic criterion showed a significant effect of orthographic consistency only in lexically-contingent shadowing. Taken together, our results suggest that lexical access is a necessary condition to the full emergence of the orthographic consistency effect. In addition, the assumption proposed by Ziegler and collaborators (2004) that the size of the consistency effect would depend on the degree of consistency of the stimuli was not confirmed in our study. Indeed, once differences in overall response latencies were partialled out, similar effect sizes were obtained in French and Portuguese, despite the difference in the degree of orthographic (in)consistency between these two languages.