To compare the efficacy and safety of didanosine and zalcitabine in patients who could not tolerate zidovudine or who had failed to respond adequately. A multicentre, randomly allocated, open-label clinical trial was set up with 230 patients treated with didanosine and 237 treated with zalcitabine. All had previously been treated unsuccessfully with zidovudine. The patients were followed for at least 1 year, with an average of 16 months. Disease progression or death occurred in 157 patients taking didanosine and 152 taking zalcitabine. There appeared to be a slight trend in favour of survival in the latter group. There is an urgent need for more effective and better tolerated antiretroviral agents in the treatment of HIV infection. Zalcitabine is at least as effective as didanosine and may provide a survival advantage in patients treated unsuccessfully with zidovudine.