OBJECTIVES: To compare the effectiveness of first protease inhibitor (PI)-containing and non-nucleoside reverse transcriptase inhibitor (NNRTI)-containing regimens. METHODS: Data were analysed from three large HIV patient databases: Apache HIV Insight (APACHE), Target Management Services (TMS) and Clinical Partners (CP). The effectiveness of therapy was the time taken for HIV-1 RNA to fall below detectable levels on first highly active antiretroviral therapy regimen (PI- or NNRTI-containing) and the subsequent time to failure (two consecutive detectable measurements). Comparisons were made using proportional hazards models, adjusting for differences in age, sex, previous reverse transcriptase inhibitor use, calendar year and baseline viral load and CD4 T-cell count. RESULTS: The type of regimen was not associated with time to undetectable viral load in any of the three databases, all of which had high power to detect a difference. PI-containing regimens were significantly less likely to fail after reaching undetectable viral loads for APACHE and CP patients (relative hazard, 1.7; 95% confidence interval, 1.3--2.1 and relative hazard, 1.6; 95% confidence interval, 1.0--2.5 respectively). These results remained significant after allowing for an unmeasured confounder with moderate effect on risk. No significant association between time to failure and regimen was found for TMS patients, possibly due to low power (67% to detect a relative hazard of 1.5). No difference was found between regimens in the time taken for an increase of > 100 x 10(9)cells/l in CD4 T-cell count. In the APACHE database, those on NNRTI-containing regimens were more likely to have a failing CD4 T-cell response. CONCLUSIONS: PI-containing regimens have a lower risk of treatment failure than NNRTI-containing regimens.