The authors evaluated 50 cases of primary melanoma of the choroid and ciliary body. In each case, 100 cells were randomly selected from a single histologic slide, and on each cell computer-assisted measurements were made of 18 nuclear and nucleolar features. Means and standard deviations were calculated for each of these features in each tumor. Thirteen calculated variables (six means and seven standard deviations) were found to correlate significantly with patient mortality following enucleation. Standard deviations of statistically significant nuclear and nucleolar features demonstrated significantly greater correlation with mortality than the means of these features, thus confirming the great value of nuclear pleomorphism for predicting the malignant potential of uveal melanomas. Furthermore, when the standard deviation of the nucleolar circumference, a feature highly correlated with survival (P less than 0.00001), was combined with the measurement of the largest dimension of the tumor, linear discriminant analysis correctly predicted the clinical course of 88 per cent of cases.