Methodological issues have limited neuroimaging studies of cerebellar structures. In this article we describe a method that addresses some of these limitations and phantom studies that examine the validity of the image manipulations. We compared volumes derived from 3D Spoiled Gradient Recalled Acquisition MR images sliced with respect to three different alignment methods: one based on cerebellar landmarks, another on cerebral landmarks and a third on the plane of acquisition. Examination of coefficients of variation, coefficients of error and convergent validity suggests that although regional cerebellar volumes based on cerebellar landmarks provide the best estimates of the true volumes, observed differences between volume measurements from alignments based on cerebellar or cerebral landmarks were generally not significant and were inconsequential. In this case, the measure was improved with alignment along local, relevant cerebellar landmarks. A set of phantom experiments showed that realignment, reslicing and interpolation in 3-dimensional image processing exerted, at most, trivial distortion on the estimates of actual object volumes.