The purpose of this study was to determine which method of functional magnetic resonance image analysis had the highest reliability in cortical and cerebellar areas during a continuous motor task. Fourteen subjects underwent 4 testing trials during 2 testing sessions separated by 3 weeks. Subjects performed a joystick task. Methods of analysis evaluated included: percent signal intensity change, active voxel count, and percent contribution. Two types of activation thresholds were evaluated: P ≤.05 and false discovery rate =.05. Reliability was determined with intraclass correlation coefficients and a repeated measure ANOVA was used to determine whether there was a significant difference between trials. A group analysis was assessed with coefficient of variation. Results indicate within session reliability was higher than between session and that signal intensity is more reliable than voxel count analysis. The intraclass correlation coefficients across different regions of interest varied depending on analysis technique. The data did not support a clear difference between thresholds. The group analysis also found less variability with intensity measures than voxel count. A stabilization effect was seen after the first trial of the experiment, suggesting that in pretest/posttest designs, a more stable result may be obtained by performing a test trial prior to actual data collection.