Since the 1990s, changing trends have been documented in species distribution and susceptibility to bloodstream infections caused by Candida species in cancer patients. However, few data are available regarding the association between in vitro antifungal susceptibility and outcome of candidemia in this patient population. We therefore evaluated the association of in vitro antifungal susceptibility and other risk factors with failure of initial antifungal therapy in cancer patients with candidemia. Candidemia cases in cancer patients from 1998 to 2001 (n = 144) were analyzed retrospectively along with their in vitro susceptibility to amphotericin B, fluconazole, and itraconazole (National Committee for Clinical and Laboratory Standards M27-A method). Patients were evaluable for outcome analysis if they received continuous unchanged therapy with either fluconazole or amphotericin B for >/=5 days. We excluded cases of mixed candidemia. In vitro susceptibility testing data of the first Candida bloodstream isolate were analyzed. Appropriate therapy was defined as that using an active in vitro antifungal for >/=5 days. For fluconazole susceptible-dose dependent Candida species, we defined appropriate therapy as a fluconazole dose of >/=600 mg/day. The Candida species distribution was 30% Candida albicans, 24% Candida glabrata, 23% Candida parapsilosis, 10% Candida krusei, 9% Candida tropicalis, and 3% other. Overall, amphotericin B was the most active agent in vitro, with only 3% of the isolates exhibiting resistance to it (>1 mg/L). Dose-dependent susceptibility to fluconazole and itraconazole was seen in 13% and 21% of the isolates, respectively, while resistance to fluconazole and itraconazole was seen in 13% and 26%, respectively.Eighty patients were evaluable for outcome analysis. In multivariate analysis, the following factors emerged as independent predictors of failure of initial antifungal therapy: leukemia (p = 0.01), bone marrow transplantation (p = 0.006), and intensive care unit stay at onset of infection (p = 0.02). Inappropriate antifungal therapy, as defined by daily dose and in vitro susceptibility, was not shown consistently to be a significant factor (it was significant in multivariate analysis, p = 0.04, but not in univariate analysis), indicating the complexity of the variables that influence the response to antifungal treatment in cancer patients with candidemia.