The prognostic significance of disease features recorded at the time of diagnosis was examined among 813 patients with Philadelphia chromosome-positive, nonblastic chronic granulocytic leukemia (CGL) collected from six European and American series. The survival pattern for this population was typical of "good-risk" patients, and median survival was 47 mo. There were multiple interrelationships among different disease features, which led to highly significant correlations with survival for some that had no primary prognostic significance, such as hematocrit. Multivariable regression analysis indicated that spleen size and the percentage of circulating blasts were the most important prognostic indicators. These features, and age, behaved as continuous variables with progressively unfavorable import at higher values. The platelet count did not influence survival significantly at values below 700 X 10(9)/liter but was increasingly unfavorable above this level. Basophils plus eosinophils over 15%, more than 5% marrow blasts, and karyotypic abnormalities in addition to the Ph1 were also significant unfavorable signs. The Cox model, generated with four variables representing percent blasts, spleen size, platelet count, and age, provided a useful representation of risk status in this population, with good fit between predicted and observed survival over more than a twofold survival range. A hazard function derived from half of the patient population successfully segregated the remainder into three groups with significantly different survival patterns. We conclude that it should be possible to identify a lower risk group of patients with a 2-yr survival of 90%, subsequent risk averaging somewhat less than 20%/yr and median survival of 5 yr, an intermediate group, and a high-risk group with a 2-yr survival of 65%, followed by a death rate of about 35%/yr and median survival of 2.5 yr.