Reading aloud is faster when targets (e.g., "PAIR") are preceded by visually masked primes sharing just the onset (e.g., "pole") compared to all different primes (e.g., "take"). This effect is known as the "masked onset priming effect" (MOPE). One crucial feature of this effect is its presumed non-lexical basis. This aspect of the MOPE is tested in the current study. Dutch participants named pictures having bisyllabic names, which were preceded by visually masked primes. Picture naming was facilitated by first-segment but not last-segment primes, and by first-syllable as well as last-syllable primes. Whole-word primes with first or last segment overlap slowed down picture naming latencies significantly. The first-segment priming effect (i.e., MOPE) cannot be accounted for by non-lexical response competition since pictures cannot be named via the non-lexical route. Instead, the effects obtained in this study can be accommodated by a "speech-planning account" of the MOPE.