We propose a modification of Wells' et. al. technique for bias field estimation and segmentation of MR images. Replacement of the class other that includes all tissue not modeled explicitly by Gaussians with small variance by a uniform probability density, and amending the EM algorithm appropriately, gives significantly better results. The performance of any segmentation algorithm is affected substantially by the number and selection of the tissue classes that are modeled explicitly, the corresponding defining parameters, and, critically, the spatial distribution of tissues in the image. We present an initial exploration of the application of minimum entropy to choose automatically the number of classes and the associated parameters that give the best output.