Neuron count in various brain structures is an important factor in many neurobiological studies. Alzheimer's disease researchers, for example, have recently suggested staging models that link clinical phenomena of the disease to characteristic neural transformations at a cellular level. An automated system for quantifying cell populations in tissue samples would improve the reliability and reduce the cost associated with obtaining cellular data. We describe a machine vision system that uses color images for the automated classification and counting of neurons in 3-D tissue samples. Samples are sliced into registered sections with thicknesses on the order of the diameter of a neuronal nucleus. Sections are stained so that the spectral transmission of neuronal nuclei differs from the surrounding tissue in preparation for imaging using a light microscope. A Bayesian classifier and geometric analysis system are employed to segment neuron regions in each section. The 3-D tissue sample is reconstructed using registered neuron regions from section to section. This allows an unbiased neuron count estimate for the sample. Preliminary experimental results are presented and compare favorably with results obtained independently by a histologist.