Neurobiology of reading research has yet to explore whether reliance on the ventral-lexical stream during word reading can be enhanced by the instructed reading strategy, or whether it is impervious to such strategies. We examined Instructions: "name all" vs. "name words" (based on spelling), Word Type: "regular words" vs. "exception words", and Word Frequency (WF) in print (log[subscript 10] HAL WF) in an experiment while measuring fMRI BOLD and overt naming reaction time (RT) simultaneously. Instructions to "name words" increased overall reliance on the ventral-lexical stream, as measured by visible BOLD activation and the WF effect on RT, with regular words showing the greatest effects as a function of this reading strategy. Furthermore, the pattern of joint effects of these variables on RT supports the notion of cascaded, not parallel, processing. These results can be accommodated by dual-stream cascaded models of reading, and present a challenge to single-mechanism parallel processing models. (Contains 5 figures.)