We tested the validity of a previously-published AIDS staging system by examining AIDS-defining diseases (ADDs) and CD4 counts as prognostic factors for survival of the 248 AIDS patients in the Edinburgh City Hospital Cohort, of whom 56% were injecting drug-users (IDUs). Cox regression was used to model the proportionality of risk of death as the CD4 count declined and more ADDs were experienced, and dependence upon post-AIDS treatment. Using the system of Mocroft et al. (Lancet 1995; 346:12-17) to grade severity, our data were well enough modelled, but we suggest: (i) regrading of HIV dementia (RR 3.9, 95% CI 2.5-6.0), mainly attributed to the drug users, to a very severe ADD; (ii) reduction in risk from zidovudine (RR 0.7, 95% CI 0.5-1.0) during AIDS follow-up for patients starting treatment at or after AIDS diagnosis; (iii) improved management of first mild ADDs (from 1987-89 to 1994-95: 40% reduction in IDUs appearing with mild index diseases, and an approximate three-fold reduction in risk associated with a mild ADD). This study supports previous findings on the significance of ADDs and lowest CD4 count in predicting the lifetime of AIDS patients.