The relation between characteristics known at start of therapy and response in 102 adults with refractory acute leukemia who received 4'-(9 acridinylamino)-methane-sulfon-m-anisidide (AMSA) was examined. Twenty-three (23%) of these patients attained complete remission (CR). Univariate analysis showed that the following characteristics were associated with CR: a fewer number of prior induction and maintenance regimens, a shorter time between latest relapse and AMSA therapy, the presence of Auer rods, a circulating blast cell count of less than 25,000/mm3, a marrow cellularity less than 90%, a ratio of marrow blasts and promyelocytes to more mature myeloid cells (differentiation ratio) less than 15, and a first-course AMSA dose of greater than or equal to 375 mg/m2. Some of these factors were interrelated. Multivariate analysis using logistic regression techniques was carried out to determine which of the above factors added independent prognostic information. This analysis produced a statistical model that related probability of response to the following in order of selection: Auer rod status, first-course dose, differentiation ratio, and absolute circulating blast cell count. Such a model could be useful in identifying patients with high, intermediate, or low probability of response to AMSA. AMSA could then be prescribed only for patients likely to respond while affording other patients alternate salvage programs at an earlier time.