In 15 normal subjects we investigated the effect on visually guided saccadic eye movements of giving a single transcranial magnetic stimulus through a circular coil centered at the vertex. In the normal paradigm, subjects fixated a target which moved randomly to the left or right by 11 degrees. The mean saccadic reaction time of 189 ms was increased by 40-50 ms if a magnetic stimulus was given in random trials some 60 ms prior to the expected onset time of control saccades. The duration and amplitude of the saccades was unchanged. The delay was smaller if the stimulus was given earlier in the reaction period, or if the coil was moved anterior or posterior to the vertex. Larger stimulus intensities produced longer delays. Three subjects were trained to produce express saccades (mean saccadic reaction times of 107-141 ms) in a 'gap' paradigm. The latency of these saccades, which are thought to be mediated by collicular mechanisms without involvement of the cortex, was not affected by magnetic stimulation. This suggests that magnetic stimulation delays normal visually guided saccades by an action on the cerebral cortex, rather than on the oculomotor centres of the brainstem. Five subjects made non-targeted saccades in darkness in response to an auditory stimulus. These saccades, like visually guided saccades, could be delayed by magnetic brain stimulation. We conclude that saccadic delay is produced by interference with cortical areas involved in the execution of saccades rather than by interfering with the perception of the visual or auditory 'go' stimulus. These probably include supplementary and frontal eye field and posterior parietal cortex. The fact that visually guided saccades emerged intact after the delay indicates that the instructions for amplitude and direction were stored separately from those involved in timing when the movement was to occur.