Experimental and clinical data implicate inadequate erythropoietin production as an important reason that infants acquire this anemia and suggest that recombinant human erythropoietin (r-HuEPO) might be used to treat or prevent it. We therefore randomly assigned 20 small premature infants (birth weight less than or equal to 1250 gm) who were highly likely to require erythrocyte transfusions for anemia of prematurity to receive 6 weeks of treatment with either intravenously administered r-HuEPO (at a dose of 100 units/kg twice each week) or a placebo. Hematologic measurements, transfusion requirements, and growth were followed during therapy and for 6 months thereafter. Treated (EPO) and control babies did not differ with respect to weight, hematocrit, overall mean absolute reticulocyte count, calculated erythrocyte mass, or rate of growth. However, reticulocyte counts increased earlier in patients given r-HuEPO. Six of ten babies in the EPO group, and 8 of 10 assigned to the control group, received at least one erythrocyte transfusion during treatment. For all infants the amount of blood sampled for laboratory tests was strongly predictive of the volume of packed erythrocytes transfused (r = 0.890; p = 0.0001). Of nine infants who had less than 20 ml packed erythrocytes removed for laboratory tests, none of four given r-HuEPO received a transfusion, whereas three of five infants assigned to the placebo group received one. No toxic effects were attributable to r-HuEPO, and no significant changes in leukocyte or platelet counts occurred during treatment. Reticulocyte counts were correlated with simultaneous platelet counts and were inversely related to absolute neutrophil counts in both study groups. We conclude that r-HuEPO administration is safe and feasible at the dose studied. Additional controlled trials utilizing higher doses of r-HuEPO and larger numbers of patients are justified.