To evaluate length and predictors of survival among children with AIDS, 529 pediatric cases diagnosed in Italy from 1983 to August 1995 were reviewed. Data were analyzed using Kaplan-Meier curves and the Cox proportional hazards regression model. Various survival patterns were subsequently analyzed. All survival analyses were truncated on March 1, 1996. Cases were examined by gender, age at diagnosis, HIV transmission category, type and number of the first AIDS-defining diseases, level of immunosuppression at AIDS diagnosis, HIV transmission category of the mother, and period of diagnosis. The overall median survival time was approximately 24 months. There wer no significant differences in survival by gender, HIV transmission category, mother's risk factor, or period of diagnosis. The Kaplan-Meier analysis showed the greatest differences in survival time between children less than 6 months of age at diagnosis (median survival 6.4 months) and all others (median 28.7 months). Children with recurrent bacterial infections or lymphoid interstitial pneumonia (LIP) had a survival time at least four times longer than those with Pneumocystis carinii pneumonia (PCP), mycobacteriosis, cytomegalovirus, tumors, or progressive multifocal leukoencephalopathy. At the multivariate analysis, the risk of death was lower among children with LIP (Relative Hazard [RH] 0.72) compared with other opportunistic diseases, whereas age less than 6 months, diagnosis of PCP or of two or more diseases, and severe immunosuppression at diagnosis increased the risk of death. Both demographic factors (age) and clinical factors (type and number of initial diseases, level of immunosuppression) were found to be independent predictors of a poor prognosis in children with AIDS. This information may be of use in improving prognosis and planning healthcare and treatment.