Concorde is a double-blind randomised comparison of two policies of zidovudine treatment in symptom-free individuals infected with human immunodeficiency virus (HIV): (a) immediate zidovudine from the time of randomisation (Imm); and (b) deferred zidovudine (Def) until the onset of AIDS-related complex (ARC) or AIDS (CDC group IV disease) or the development of persistently low CD4 cell counts if the clinician judged that treatment was indicated. Between October, 1988, and October, 1991, 1749 HIV-infected individuals from centres in the UK, Ireland, and France were randomly allocated to zidovudine 250 mg four times daily (877 Imm) or matching placebo (872 Def). Follow-up was to death or Dec 31,1992 (total 5419 person-years; medians 3·3 years) and only 7% of the 1749 had not had a full clinical assessment after July 1, 1992. Of those allocated to the Def group, 418 started zidovudine at some time during the trial, 174 (42%) of them at or after they were judged by the clinician to have developed ARC or AIDS (nearly all confirmed subsequently) and most of the remainder on the basis of low CD4 cell counts. Those in the Imm group spent 81% of the time before ARC or AIDS on zidovudine compared with only 16% for those in the Def group. Despite the large difference in the amount of zidovudine between the two groups and the fact that the number of clinical endpoints (AIDS and death) in Concorde (347) outnumbers the total of those in all other published trials in symptom-free and early symptomatic infection, there was no statistically significant difference in clinical outcome between the two therapeutic policies. The 3-year estimated survival probabilities were 92% (95% Cl 90-94%) in Imm and 94% (92-95%) in Def (log-rank p=0·13), with no significant differences overall or in subgroup analyses by CD4 cell count at baseline. Similarly, there was no significant difference in progression of HIV disease: 3-year progression rates to AIDS or death were 18% in both groups, and to ARC, AIDS, or death were 29% (Imm) and 32% (Def) (p=0·18), although there was an indication of an early but transient clinical benefit in favour of Imm in progression to ARC, AIDS, or death. However, there was a clear difference in changes in CD4 cell count over time in the two groups. Median changes from baseline at 3 months were + 20 cells/µL (Imm) and -9 cells/µL (Def), a difference of 29 cells/µL (95% Cl 16-42; p < 0·0001) which persisted for up to 3 years. Thus such persistent differences in CD4 cell count do not necessarily imply long-term differences in clinical outcome. 6 participants had life-threatening adverse events that were judged to be possibly drug related: 4 occurred on trial capsules before unblinding (3 on zidovudine, 1 on placebo) and 2 occurred on open zidovudine. Despite a daily dose of 1 g of zidovudine, the frequency of severe haematological and other adverse events on trial therapy was low but significantly higher in the Imm group: 16 Imm and 2 Def participants stopped trial drug for haematological events and the estimated proportions with haemoglobin dropping below 10 g/dL were 5% and 1%, respectively, at 1 year and 8% and 2%, respectively, at 3 years. Another 83 (9%) Imm and 36 (4%) Def participants stopped for other adverse events, predominantly gastrointestinal or neurological symptoms (headaches) or malaise. The results of Concorde do not encourage the early use of zidovudine in symptom-free HIV-infected adults. They also call into question the uncritical use of CD4 cell counts as a surrogate endpoint for assessment of benefit from long-term antiretroviral therapy.