One hundred and six patients aged </=60 years with newly diagnosed acute myeloid leukaemia (AML) treated with fludarabine-based regimens (cases) were matched with 106 AML patients treated with conventional non-fludarabine-based regimens (controls). The cases and controls were matched by expression of the multidrug resistance P-glycoprotein (MDR-Pgp), measured by flow cytometry as mean fluorescence index (MFI), cytogenetics, and age. The complete remission (CR) rate of the cases was 61% among the MDR-Pgp-positive (pos(ve)) patients (MFI >/= 6) vs. 75% among the MDR-Pgp-negative (neg(ve)) ones (MFI < 6) (P = 0.16). Conversely, in the controls, the CR rate was 44% among the MDR-Pgp-pos(ve) patients vs. 67% among the MDR-Pgp-neg(ve) ones (P = 0.02). The 4-year disease-free survival (DFS) and overall survival (OS) of MDR-Pgp-pos(ve) cases were significantly longer than those of MDR-Pgp-pos(ve) controls (DFS, 28.1% vs. 6.5%, P = 0.004; OS, 33.5% vs. 9.6%, P = 0.01). This difference was not found among the MDR-Pgp-neg(ve) patients. By univariate (P = 0.007) and multivariate (P = 0.007) analysis, the MDR-Pgp-pos(ve) phenotype was negatively correlated with CR and it emerged as the most important independent negative prognostic factor, after cytogenetics. Our study confirms the prognostic impact of the MDR phenotype in AML and strongly suggests fludarabine-based induction treatments as a promising strategy for MDR-Pgp-pos(ve) AML patients. In this setting of patients, large prospective randomised studies should be planned.