Reading is a highly complex task involving a precise integration of vision, attention, saccadic eye movements, and high-level language processing. Although there is a long history of psychological research in reading, it is only recently that imaging studies have identified some neural correlates of reading. Thus, the underlying neural mechanisms of reading are not yet understood. One very practical requirement of reading is that eye movements be precisely controlled and coordinated with the cognitive processes of reading. Here we present a biologically realistic model of the frontal eye fields that simulates the control of eye movements in human readers. The model couples processes of oculomotor control and cognition in a realistic cortical circuit of spiking neurons. A global rule that signals either "reading" or "not reading" switches the network's behavior from reading to scanning. In the case of reading, interaction with a cortical module that processed "words" allowed the network to read efficiently an array of symbols, including skipping of short words. Word processing and saccade buildup were both modeled by a race to threshold. In both reading and scanning, the network produces realistic distributions of fixation times when compared with human data.