Caspofungin is a new antifungal drug of the echinocandin class. We analyzed the clinical efficacy of caspofungin (50 mg/day) in the treatment of HIV-infected adults with endoscopically documented Candida esophagitis and enrolled in four clinical trials of caspofungin. Symptoms were evaluated daily; a favorable outcome required complete resolution of all esophageal symptoms assessed at the time of discontinuation of therapy. Relapse was defined as recurrent symptoms during the subsequent 2 weeks. A multivariate logistic regression model was developed to identify potential factors (including severity of symptoms at presentation, CD4(+) cell count on entry, extent of disease [assessed endoscopically at baseline], causative Candida species, duration of therapy [overall and after resolution of symptoms], time on treatment before symptom resolution, and antifungal prophylaxis) that might predict symptomatic relapse in the 2 weeks following completion of therapy. The median CD4(+) lymphocyte count for the entire population was 31/mm(3). Candida albicans was isolated from 109 of 110 patient samples cultured for the pathogen and constituted the sole isolate in 77%. Extensive esophageal involvement was present in 55% of patients at the time of pretreatment endoscopy. The duration of therapy ranged from 7 to 20 days (median, 12 days). Symptoms resolved in 117 of 123 patients (95%; 95% confidence interval, 90-98%) with a median time of ~4 days. Response rates were 43 of 46 (93%) and 70 of 73 (96%) for patients with greater or fewer than 50 CD4(+) cells/mm(3), and 80 of 85 (94%) and 23 of 24 (96%) in infections caused by C. albicans alone or in association with non-albicans isolates, respectively. Symptoms recurred within 2 weeks of stopping caspofungin in 19 of 115 evaluable patients (17%), including 3 of 16 (19%) receiving antifungal prophylaxis. Relapse rates were similar for patients with greater or fewer than 50 CD4(+) cells/mm(3). In this relatively small number of patients, only symptom severity and extent of disease judged endoscopically at baseline were significantly (p < 0.10) associated with early relapse in the multivariate model.