Candidemia is a clinically important disease which has increased in incidence worldwide in recent decades. In order to identify the risk factors for mortality in candidemic patients and to elucidate the role of antifungal susceptibility testing, a retrospective cohort study was performed of 56 episodes of candidemia in 1998 at a medical center in southern Taiwan. The minimal inhibitory concentration (MIC) of these isolates was determined by E-test. Malignancy and alimentary diseases (42.9%) were the most common underlying conditions of these patients. There was no difference of Candida spp. distribution among patients treated in medical or surgical departments, except that all 5 isolates of C. intermedia were found in patients treated in medical departments (p=0.02) and 50% of candidemic infants had C. parapsilosis isolates (p=0.046). Among all Candida isolates, 3 (5.4%) were fluconazole non-susceptible. C. tropicalis had a significantly higher rate of amphotericin B resistance than the other species (p=0.007). Thirty four patients died and 70.6% of these deaths were attributable to candidemia. Thrombocytopenia, septic shock at the date of candidemia onset, C-reactive protein > 100 mg/L, blood urea nitrogen > 20 mg/dL, length of stay < 60 days, and Acute Physiology and Chronic Health Evaluation II score > or = 10 points were significantly associated with the death attributable to candidemia. Thrombocytopenia was the only independent predictor for mortality in the multivariate analysis. When the breakpoint of fluconazole was set at 2 microg/mL, as opposed to 8 microg/mL as in the National Committee for Clinical Laboratory Standards (NCCLS) criteria, the clinical outcome of death was significantly correlated to the MICs of the blood isolates. The correlation between MIC of fluconazole determined by E-test data, which is more easily obtainable than with NCCLS methods, and outcome requires larger scale investigation.