Poor adherence to HIV protease inhibitors may compromise the effectiveness of treatment. Few studies have compared methods for measuring adherence or have related adherence measures to a clinical outcome. To examine the relationship among a composite score of adherence, the three primary measures of adherence, and HIV virologic response. Longitudinal cohort study. Public HIV clinic. 108 HIV-infected adults receiving protease inhibitors or non-nucleoside reverse transcriptase inhibitors who were monitored for 666 monthly intervals. Medication Event Monitoring System (MEMS), pill count, and interview combined into a composite adherence score (CAS), and HIV viral load. Mean antiretroviral adherence differed by adherence measure (MEMS, 0.63; pill count, 0.83; interview, 0.93; and CAS, 0.76). Composite adherence score decreased significantly over time. Composite adherence score, MEMS values, pill values, and interview values were statistically significantly associated with achievement of an undetectable viral load within 6 months of initiating therapy. Composite adherence score showed the strongest predictive relationship (odds ratios for a 10% increase in adherence for CAS, MEMS, pill count, and interview, respectively, were 1.26 [95% CI, 1.16 to 1.37], 1.13 [CI, 1.06 to 1.21], 1.10 [CI, 1.02 to 1.19], and 1.35 [CI, 0.94 to 1.94]). Different measures applied to the same patient suggest different levels of adherence. Adherence may be underestimated by MEMS and overestimated by pill count and interview. A summary measure combining several measures is more strongly related to a clinical response, but more practical measurement methods are needed for clinical use.