In previous studies, the prognostic value of bone marrow (BM) histologic patterns in chronic lymphocytic leukemia (CLL) has been demonstrated. In order to investigate whether such a value is independent of other prognostic parameters, a multivariate survival analysis (Cox's regression model) was undertaken in a series of 329 CLL patients in whom a BM had been performed. The following binary variables were included in the analysis: age (more than 60 years), lymphadenopathy (more than two areas involved), splenomegaly, hepatomegaly, absolute lymphocyte count (more than 30,000 microL), anemia (hemoglobin less than 10 g/dL), thrombocytopenia (less than 100,000 microL), and BM pattern (diffuse v nondiffuse). Three variables entered the regression at significant level: BM pattern (P less than .001), anemia (P less than .001), and hepatomegaly (P = .03). The model was also tested by expressing the variables in a continuous way when possible. Again, BM pattern entered first in the regression (P less than .001), followed by the hepatomegaly (P = .002), hemoglobin level (P = .02), and lymphadenopathy (P = .04). When both the binary and the continuous models were tested separately in 227 patients with BM as initial staging procedure and in 102 patients in whom this was performed later during the course of the disease, in all instances, BM pattern entered first in the regression at a highly significant level. BM histologic pattern appears to be a better single prognostic parameter than any one of the variables employed in current clinical staging systems. A combined clinicopathologic system incorporating the BM pattern, together with the usual clinical variables, is presented.