Within-subject reproducibility of visual brain activation using multislice echo planar functional magnetic resonance imaging (fMRI) was tested. Ten healthy subjects underwent fMRI with visual stimulation on three occasions: two studies in one scanning session (without repositioning); and a third study 1 h to 2 weeks later. Following a three-dimensional matching procedure, activation was measured and compared between sessions on a voxel-by-voxel basis. Data were filtered to full-width-at-half-maximum of 4.0 x 4.0 x 5.0 mm and a conservative Bonferroni-corrected significance threshold was applied to correlation maps. For reproducibility, change in centre of mass of the activated volume, a ratio of the number of pixels and a ratio of the number of overlapping pixels was calculated. Further, reproducibility was tested varying significance thresholds and at different filter widths. Average changes in centre of mass of the activated volume were 2.63 and 3.96 mm between Studies 1 and 2 and 1 and 3, respectively. The reproducibility of the number of activated voxels was 90% and 88% (Studies 1 and 2 and 1 and 3). The ratio of overlapping pixels was 74% between Studies 1 and 2 and 64% between Studies 1 and 3. Varying the significance threshold showed that at a certain range, the overlap reached a maximum, and increasing the filter widths increased reproducibility. It is concluded that fMRI with visual stimulation can be used to measure brain activity with reasonably good reproducibility on a routine clinical system equipped with echo planar imaging. Difficulties remain in separating the contribution of motion, repositioning errors, and true physiological changes.