From the start of clinical testing for human immunodeficiency virus (HIV) antibody in 1985 until the end of June 1994, the Microbiological Diagnostic Unit tested 134 346 serum samples from 109 391 individuals, identifying 672 as infected. A suitable database, using coded identities, was established from the beginning. Because the sera came from a variety of clinical settings, it was possible to extract information about patterns of requests for tests and about the yield of seropositive cases. The data were able to highlight factors associated with high rates of positive tests that can be lost in amalgamated statewide data. The discovery of HIV-positive females was sporadic and there was a much lower detection rate than from testing males. While males were most commonly diagnosed through specialist venereology services, 61 per cent of diagnoses in females were through nonspecialised practices, often those without previous experience in diagnosis and management of HIV infection. Despite some individual exceptions, the requesting of HIV tests appears generally to have been justifiable (particularly in view of the direction of official preventive campaigns) and not profligate. Overall, where practice was unsatisfactory, this lay less in thoughtless ordering of tests than in failing to include with the specimen the elementary, but vital, epidemiological information requested.