Response to salvage therapy at first and second relapse was analyzed in 150 patients with acute myeloid leukemia (AML) to improve the characterization of relapsed AML and to deduce from this analysis a proposal for the definition of refractoriness against conventional therapy. Salvage treatment consisted of a repetition of the TAD 9 regimen which was already applied as induction protocol at initial diagnosis. All patients were recruited from the multicenter 1982 trial of the German AML Cooperative Group and had thus received a standardized first line treatment. Response at first relapse was significantly related to the duration of the first remission. From 38 patients relapsing within 6 months after successful induction therapy, only 11 (28%) achieved a second complete remission as compared to 58 of 98 (59%) cases with later occurring relapses (p less than 0.01). This difference was due to a significantly higher incidence of persistent leukemia in the former group and not biased by differences in early death rates. No other variable was found predictive for the response to salvage treatment including age, WBC, serum LDH, morphologic subtype, presence or absence of DNA aneuploidy as detected by flow cytometry or maintenance chemotherapy. A low remission rate of 28% was also obtained in the 14 patients at second relapse. These data indicate that patients with a duration of their first remission of more than 6 months cannot be considered as being refractory against standard chemotherapy while patients with early relapses and second recurrences have a response rate of less than 30% due to refractory disease. Hence, the following criteria are proposed for the definition of refractoriness against standard chemotherapy in advanced AML: (a) nonresponse to first-line induction therapy, (b) early relapse within 6 to 12 months of first remission, (c) relapse after 6 to 12 months of first remission and failure on a reinduction attempt with established regimens, (d) second and subsequent relapses. These criteria may provide a useful rationale for the selection of the most appropriate treatment at relapse. They may also serve as eligibility criteria for clinical phase I/II studies and will facilitate interstudy comparisons.