The center of mass (COM) in functional MRI studies is defined as the center of a cerebral activation cluster. Although the COM is a well-accepted parameter for exactly localizing brain function, the reliability of COMs has not received much attention until now. Our goal was to investigate COM reliability as a function of the thresholding technique, the threshold level, and the type of COM calculation. Therefore 15 subjects were examined repeatedly using simple hand and tongue movement paradigms. Postprocessing was performed with uncorrected, corrected, and proportional thresholding as well as different threshold levels. Geometric and T-weighted COMs of left-hemispheric primary hand and tongue motor clusters were calculated. The COM variation was evaluated within and between repeated sessions depending on the different postprocessing setups. Mean COM variations over three repeated sessions varied between 1.6 mm and 9.8 mm for the hand paradigm and between 7.0 mm and 14.4 mm for the tongue task. Stringent thresholding techniques and high threshold levels were required to assess reliable results, whereas the kind of COM calculation was of lesser relevance. Thus, COM reliability cannot be presupposed; it depends strongly on the individual postprocessing techniques. This should be considered when using COMs for localizing brain function.