This paper provides the first comprehensive report on mortality by type of leukemia among the Japanese atomic bomb survivors in the Life Span Study (LSS). Analyses include 310 deaths due to leukemia during the period 1950-2000 among 86,611 people in the LSS. Poisson regression methods were used to evaluate associations between estimated bone marrow dose and leukemia mortality. Attention was given to variation in the radiation dose-leukemia mortality association by time since exposure, age at exposure, city and sex. The excess relative rate per gray of acute myeloid leukemia was best described by a quadratic dose-response function that peaked approximately 10 years after exposure. Acute lymphatic leukemia and chronic myeloid leukemia mortality were best described by a linear dose-response function that did not vary with time since exposure. Adult T-cell leukemia was not associated with estimated bone marrow dose. Overall, 103 of the 310 observed leukemia deaths were estimated to be excess deaths due to radiation exposure. In the most recent decade of observation (1991-2000), the estimated attributable fraction of leukemia deaths among those survivors exposed to >0.005 Gy was 0.34, suggesting that the effect of the atomic bombings on leukemia mortality has persisted in this cohort for more than five decades.