In order to develop a diagnostic approach to the common problem of anemia associated with alcoholism, 121 chronic alcoholics admitted to a general medical service with a low hematocrit were evaluated. Multiple contributing causes of anemia were present in most patients. Megaloblastic marrow change was found in 33.9% of patients, sideroblastic change in 23.1%, absent iron stores in 13.2%, aggregated macrophage iron in 81.0%, and acute blood loss in 24.8%. The MCV was of little value in predicting the presence of megaloblastic change unless markedly elevated (greater than 110 fl). In 15 of 41 patients with megaloblastic marrow morphology (36.6%) the MCV was normal or low. Among 40 patients with MCV values between 100 and 110 fl, megaloblastic change was not present in the bone marrow smears of 24 (60.0%). Neutrophil hypersegmentation was 95% specific but only 78% sensitive for megaloblastic change; in contrast, the presence of macroovalocytosis was 90% sensitive but only 68% specific. Serum lactic dehydrogenase, plasma folate, and erythrocyte folate levels had such low sensitivities and specificities for megaloblastic change as to be of little predictive value. Hematologic responses to folic acid were often inadequate in patients with megaloblastic morphologic changes, apparently because of associated acute and chronic illness. Our findings are consistent with the hypothesis that 2 mechanisms account for the development of megaloblastic hematopoiesis in alcoholics: induction of folate deficiency and a direct toxic effect of alcohol on erythroid precursors independent of folate depletion, as reflected by the presence of normal plasma and erythrocyte folate levels in several patients with megaloblastic change. In no patient was sideroblastic change the sole apparent cause of anemia. Megaloblastic hematopoiesis and aggregated macrophage iron frequently accompanied sideroblastic change. Examination of the blood smear revealed siderocytes in one-third of patients with sideroblastic marrows and dimorphic erythrocyte morphology in the majority. Dimorphic blood smears, however, were neither sensitive nor specific for sideroblastic change. Serum iron concentrations were usually not elevated in the group with sideroblastic abnormalities. In predicting marrow iron stores, serum iron and iron-binding capacity concentrations were often non-diagnostic or misleading. Serum ferritin levels less than 100 ng/ml, however, showed 100% sensitivity and 95% specificity for absent marrow iron stores despite the frequent presence of abnormal liver function. On the basis of our findings, practical guidelines have been formulated for the evaluation and therapy of anemia in alcohol