Two contrasting theories of reading are reviewed in light of recent research in psycholinguistics. A strictly "visual" model of fluent reading is contrasted with several mediational theories where auditory or articulatory coding is deemed necessary for comprehension. Surveying the research in visual information processing, oral reading, subvocalization, and reading speed, the author concludes that the "visual" hypothesis is not commensurate with these findings. In particular, an experiment predicting the pause structure in fluent reading is reported, and a level of understanding involving both deep and surface structure information is proposed as characterizing such performances. Implications for beginning reading are drawn. (Author/MS)