During oral reading we hypothesized that lexical representations are activated and selected for output by the simultaneous activation of the semantic, the direct lexical orthography to phonology, and sublexical grapheme-to-phoneme conversion (GPC) routes (Southwood and Chatterjee, 1999). Serial models of reading argue that the semantic route governs oral reading with minimal influence from the nonlexical direct route the sublexical GPC route. These models predict that semantic errors should occur in reading when the semantic route and GPC are both impaired. The Simultaneous Activation Hypothesis predicts few semantic errors in oral reading many during picture naming. Semantic errors are infrequent in reading because information from all three reading routes constrains activation of a phonological entry. By contrast phonological selection in picture naming is constrained primarily by the semantic route and if damaged additional information is unavailable to select the appropriate phonological code. In agreement with the Simultaneous Activation Hypothesis five phonological dyslexics produced semantic errors during picture naming but not when reading aloud. Phonological errors were present during oral reading and minimal during picture naming.