BACKGROUND: Western blot (WB) criteria in epidemiological studies in Africa exhibit an unacceptably high proportion of indeterminate results. New diagnostic criteria are urgently needed. METHODS: From 1989 to 1998, WB confirmatory tests were performed after weakly positive or discordant results of two enzyme immunoassays in a large Ugandan population. Enzyme immunoassays (EIA) on new sera taken prospectively from the same individuals one year later were used to assess the human immunodeficiency virus (HIV) status of these people. A logistic model was used to determine which set of WB bands was the most predictive of HIV status. Diagnostic criteria were then established, based on the likely HIV status determined using the predictive values and the intensity of the bands. RESULTS: Using 1109 WB tests, the best diagnostic criteria were based on only two bands (gp160 and p31). These criteria were validated on an independent sample of 587 WB tests, giving a high sensitivity and specificity (90.3% and 97.0%, respectively) and few indeterminate results (2.7%). These criteria classified correctly 96.3% of the sera. CONCLUSION: Our diagnostic criteria gave far better results in our population than the existing published criteria. This suggests that new criteria could be developed to improve WB interpretation in African settings.