To report the occurrence of HIV-related mucosal candidosis that fails to respond to fluconazole, to establish the correlation between in vitro susceptibility testing and clinical failure, and to assess the efficacy of alternative treatments. Chart review of all patients with fluconazole failure and all patients with CD4 counts < 50 x 10(6)/l continuing to respond to fluconazole, and prospective in vitro susceptibility testing of Candida. A regional treatment centre for HIV-infected individuals in north-west England. A cohort of 155 HIV-positive individuals with CD4 counts < 300 x 10(6)/l cells. Clinical fluconazole failure was defined as symptomatic oropharyngeal or oesophageal candidosis despite fluconazole > or = 100 mg per day for 10 days. In vitro susceptibility to fluconazole was determined for Candida isolates. Cumulative 12-month fluconazole dose and time from first fluconazole therapy and prophylaxis were recorded. Nine (5.8%) patients meeting the definition of fluconazole failure were identified. In vitro susceptibility to fluconazole of temporally related oropharyngeal isolates was reduced in all cases. Intravenous amphotericin B was the only effective treatment for these patients when symptoms were severe suggesting azole cross-resistance. One patient, who had received alternative treatments for 9 months, reverted from in vitro and clinical fluconazole sensitivity but relapsed within 6 weeks of resuming fluconazole. The median fluconazole dose over the preceding 12 months for the eight adult cases was 386 mg weekly. The median dose for the same period was 79 mg weekly in 28 patients with CD4 counts < 50 x 10(6)/l but without fluconazole failure (difference, 307; 95% confidence interval, 199-514; P < 0.0001). A substantial problem of clinical fluconazole failure has developed among HIV-positive patients who have recurrent problematic mucosal candidosis.