Although there are wide differences in prognosis between patients with AIDS they are often thought of as a single homogeneous group. We think a simple staging system that accounts for important prognostic factors including type and number of AIDS diseases and the CD4 lymphocyte count is required. We followed 363 AIDS patients at the Royal Free Hospital and reported the occurrence of 680 AIDS-defining diseases (ADDs). We measured CD4 counts at approximately monthly intervals. Severity of AIDS diseases was defined a priori on the basis of survival in the AIDS in Europe study of 6578 AIDS patients: mild-oesophageal candidiasis, Kaposis sarcoma (cutaneous), Pneumocystis carinii pneumonia, extrapulmonary tuberculosis; severe-all other ADDs except lymphoma; very severe-lymphoma. The risk of death increased by 15% (p = 0.08) for each mild condition experienced, by 89% (p < 0.0001) for each new severe condition and by 535% (p < 0.0001) when a lymphoma developed. Estimates from the Cox model were used to derive a score reflecting the risk of death. Patient experience was divided into three categories. Patients in AIDS Grade I had an average death rate of one per 10.1 years, compared with one per 2.8 years in AIDS Grade II and one per 1.1 years in AIDS Grade III. Similar rates were seen in an independent validation study on 1230 AIDS patients at different hospital. Our grading system should be useful for patient management, clinical trial design, surveillance, and resource management.