BACKGROUND There have been few evaluations of the risk of acute nonlymphocytic leukemia (ANLL) following therapy for non-Hodgkin's lymphoma (NHL). Further, the relationship between cumulative dose of cytotoxic drug, radiation dose to active bone marrow, and the risk of ANLL following NHL have not been well described. OBJECTIVE Our purpose was to examine the risk of ANLL in relationship to all prior treatment for NHL. METHODS Within a cohort study of 11,386 2-year survivors of NHL, 35 case patients with secondary ANLL were identified and matched to 140 controls with NHL who did not develop ANLL. The primary eligibility criteria for the cohort included a diagnosis of NHL as a first primary cancer from January 1, 1965, through December 31, 1989; age 18 through 70 years at the time of initial diagnosis; and survival for 2 or more years without the development of a second invasive primary malignancy. Detailed information on chemotherapeutic drugs and radiotherapy was collected for all patients. Standard conditional logistic regression programs were used to estimate the relative risk (RR) of ANLL associated with specific therapies by comparing the exposure histories of case patients with individually matched controls. RESULTS Significant excesses of ANLL followed therapy with either prednimustine (RR = 13.4; 95% confidence interval [CI] = 1.1-156; P trend for dose < .05) or regimens containing mechlorethamine and procarbazine (RR = 12.6; 95% CI = 2.0-79; P < .05). Elevated risks of leukemia following therapy with chlorambucil were restricted to patients given cumulative doses of 1300 mg or more (RR = 6.5; 95% CI = 1.6-26; P < .05). Cyclophosphamide regimens were associated with a small, nonsignificant increased risk of ANLL (RR = 1.8;95% CI = 0.7-4.9), with most patients receiving relatively low cumulative doses (< 20,000 mg). Radiotherapy given at higher doses without alkylating agents was linked to a nonsignificant threefold risk of ANLL compared with lower dose radiation or no radiotherapy. CONCLUSIONS Our results suggest that prednimustine may be a human carcinogen, with a positive dose-response gradient evident for ANLL risk. The low, nonsignificant risk of leukemia associated with cyclophosphamide was reassuring because this drug is commonly used today. Despite the excesses of ANLL associated with specific therapies, secondary leukemia remains a rare occurrence following NHL. Of 10,000 NHL patients treated for 6 months with selected regimens including low cumulative doses of cyclophosphamide and followed for 10 years, an excess of four leukemias might be expected.