In children with ALL, the appearance of leukemic cells in stained spreads can be useful in predicting outcome of the disease. Cell size and nucleolar size and prominence are the two criteria, detectable on Romanowsky-stained bone marrow spreads, which permit such prediction. In a series of 93 children with ALL, 48 "poor risk" patients were detected. More than 1% of the bone marrow blast cells of these patients exceeded in diameter twice the diameter of nearby red blood cells, and, on average, more than one nucleolus was visible in every two leukemic cells. The survival of these 48 patients (estimated median 18 mo) was significantly worse (p = 0.0036) than that of the other 45 patients (estimated median over 5 yr). Analysis of the series with respect to other risk factors (age, leukocytosis at diagnosis) revealed the cytomorphological risk to be independent of the other two. Comparison of this method with others which have been used suggests that the proposed method may be more reproducible (capable of being learned) than others because it relies on specific characteristics rather than on total recognition of a "cell type," and that it may select a higher proportion of "poor risk" patients than other methods. PAS-positivity may be a useful additional criterion.