Previous studies have shown that the BOLD functional MRI (fMRI) signal is increased in several cortical areas when subjects perform anti-saccades compared with pro-saccades. It remains unknown, however, whether this increase is due to an increased cortical motor signal for anti-saccades or due to differences in preparatory set between pro- and anti-saccade trials. To address this question, we measured event-related fMRI in a paradigm that allowed us to separate instruction-related brain activity from saccade-related brain activity. In this paradigm, the instruction to either generate a pro-saccade or an anti-saccade was conveyed by a switch in the color of the central fixation stimulus and preceded the presentation of a peripheral stimulus by either 6, 10, or 14 s. Cortical areas were functionally mapped using the general linear model comparing standard pro- and anti-saccade blocks with fixation blocks. When the trials were aligned on the onset of the instruction stimulus, bilateral frontal eye fields and right hemisphere dorsolateral prefrontal cortex showed an increased signal during the instruction period on anti-saccade trials as compared with pro-saccade trials. When the trials were aligned on the movement stimulus and the instruction period activity was subtracted, there were no differences between pro- and anti-saccades. This finding suggests that the increased cortical activation found in previous blocked designs originates predominately from differences in preparatory set and not from differences in the motor signal between pro- and anti-saccades.