The objective of this study was to test the hypothesis that HIV interacts with human papilloma virus (HPV) to increase the odds of cervical neoplasia. The study design was a meta-analysis using data pooled from published sources. Studies published between January 1986 and March 1998 were eligible for inclusion if they included data on neoplasia (cytology-based), HIV (defined by laboratory and/or standard clinical criteria), and HPV (assessed by PCR, Southern blot, dot-blot hybridization, or cytology of an otherwise well designed study) among nonpregnant women. Blinded data abstraction was performed independently by the investigators. There were 15 studies that were eligible and presented data in a format that could be abstracted for analysis. Data were pooled using a Mantel-Haenszel summary odds ratio (OR); generalized estimation regression equations were used to examine independent effects of HIV and HPV. Overall, based on the Mantel-Haenszel ORs, there was a strong overall association between HPV and neoplasia [OR, 8.1; 95% confidence interval (CI), 6.5-10.1]. Stratifying by HIV status, HIV-positive women had higher odds of disease (OR, 8.8; 95% CI, 6.3-12.5) than HIV-negative women (OR, 5.0; 95% CI, 3.7-6.8). In the regression model, there was an interaction between HPV and HIV (P = 0.01); immunosuppression also tended to predict neoplasia (P = 0.058). HIV seems to be a cofactor in the association between HPV and cervical neoplasia; this effect may vary by level of immune function. These speculations are biologically plausible. Additional data from large, well designed studies are needed to confirm these hypotheses.