By several accounts, reading single words may be accomplished either by sequentially transcribing orthographic units into their corresponding sounds (an indirect route), or by directly associating a visual word form to the semantic or articulatory representation (a direct route). In contrast, the similar task of naming objects must rely only on a direct route, since objects cannot be "sounded out". To the study the localization of cognitive processes specific to reading, we used Positron Emission Tomography to measure regional cerebral blood flow while subjects named words and pictures of objects silently or aloud. Group averages of blood flow changes were obtained for experimental vs. control tasks. Object and word presentations elicited similar blood flow increases in extra-striate visual cortices compared to a visual noise control. Silent reading invoked a neural network very similar to that seen when subjects named objects silently, consistent with a "direct" route. Naming objects aloud produced the addition of motor output regions to this network. In contrast, oral reading produced a markedly different pattern of activated regions, suggesting reliance on a separate phonological pathway. These results provide support for the dual coding hypothesis in reading, and challenge the use of strict hierarchical models of cognitive operations in PET activation studies.