Microscope images of thin marble sections can be used to determine their geographical origin by means of the shape, size, and spatial distribution of their grains. We present a method that is the first step toward the automatic origin determination, namely, the segmentation of grains in digital images of thin marble sections. Each grain has a preferred direction, different from that of its neighbors, which rules its behavior when it is illuminated with polarized light. First, we perform an oversegmentation of the image with a filter based on a partial wavelet reconstruction followed by a watershed transformation. Afterward, from a sequence of images of the same sample obtained by means of polarized light we compute, for each region, two parameters that depend on the preferred direction of that region’s grain. Finally, the set of parameter values for all the regions is the input to a region-merging procedure that achieves the final segmentation. We present the results for samples from six quarries, each with different visual features.