Despite low sensitivity (around 60%), cytomorphologic examination of urine specimens represents the standard procedure in the diagnosis and follow-up of bladder cancer. Although color is information-rich, morphologic diagnoses are rendered almost exclusively on the basis of spatial information. We hypothesized that quantitative assessment of color (more precisely, of spectral properties) using liquid crystal-based spectral fractionation, combined with genetic algorithm-based spatial analysis, can improve the accuracy of traditional cytologic examination. Images of various cytological specimens were collected every 10 nm from 400 to 700 nm to create an image stack. The resulting data sets were analyzed using the Los Alamos-developed GENetic Imagery Exploitation (GENIE) package, a hybrid genetic algorithm that segments (classifies) images using automatically 'learned' spatio-spectral features. In an evolutionary fashion, GENIE generates a series of algorithms or 'chromosomes', keeping the one with best fitness with respect to a user-defined training set. First, we tested the system to determine if it could recognize malignant cells using artificial cytology specimens constructed to completely avoid the requirement for human interpretation. GENIE was able to differentiate malignant from benign cells and to estimate their relative proportions in controlled mixtures. We then tested the system on routine cytology specimens. When targeted to detect malignant urothelial cells in cytology specimens, GENIE showed a combined sensitivity and specificity of 85 and 95%, in samples drawn from two separate institutions over a span of 4 years. When trained on cases initially diagnosed as 'atypical' but with unequivocal follow-up by biopsy, surgical specimen or cytology, GENIE showed efficiency superior to the cytopathologist with respect to predicting the follow-up result in a cohort of 85 cases. We believe that, in future, this type of methodology could be used as an ancillary test in cytopathology, in a manner analogous to immunostaining, in those situations when a definitive diagnosis cannot be rendered based solely on the morphology.