Color nonstandardness - the propensity for similar objects to exhibit different color properties across images - poses a significant problem in the computerized analysis of histopathology. Though many papers propose means for improving color constancy, the vast majority assume image formation via reflective light instead of light transmission as in microscopy, and thus are inappropriate for histological analysis. In this work, we present a novel Bayesian color segmentation algorithm for histological images that is highly robust to color nonstandardness; this algorithm employs a unique instantiation of the expectation maximization (EM) algorithm to dynamically estimate - for each individual image - the probability density functions (mixtures of gamma and von Mises distributions) that describe the colors of salient objects. To validate our segmentation scheme, we employ it as part of a computerized system to detect myelodysplastic syndromes (MDS) on bone marrow specimens. Qualitative anecdotal evidence suggests that biopsies of MDS exhibit abnormalities in the arrangement of erythroid precursors (immature red blood cells). Herein, we confirm and quantify this phenomenon, using it to discriminate MDS from normal tissue: over a dataset of 53 representative regions selected from 18 patients, our classification system correctly discriminates MDS from normal tissue with an accuracy of 85% and an area under the receiver operator characteristic curve of 0.8803.