To investigate the frequency of a first therapy interruption (TI) > or = 12 weeks, to identify the factors associated with TI and with therapy resumption, and to compare the risk of developing clinical events during TI and during continuous therapy. Observational study of 3142 patients who started a first highly active antiretroviral therapy (HAART) regimen. End points were time to (1) first TI of > or = 12 weeks, (2) subsequent therapy resumption, and (3) development of new AIDS-related events or death. Over a median follow-up period of 41 months (interquartile range: 18-60 months), 721 patients (22.9%) interrupted HAART for > or = 12 weeks, with a probability of 28.6% (95% confidence interval [CI]: 26.7-30.6) by 4 years from the date of therapy initiation. Patient decision (47.4%) and toxicity (24.0%) were the main reasons for TI. Women, injection drug users, and patients with a higher current CD4 cell count were more likely to interrupt. The median time to therapy resumption was 12 months (95% CI: 11-14). The higher the current CD4 count, the slower was the rate of resuming therapy; conversely, patients who stopped because of failure and those with a pre-HAART viral load >100,000 copies/mL resumed therapy sooner. Two hundred eighty-one patients experienced clinical progression at a rate of 2.6 per 100 person-years (pys) (95% CI: 2.3-3.0) while patients were on therapy and 3.5 per 100 pys (95% CI: 2.4-4.8) during TI. The adjusted relative hazard of clinical progression associated with TI was 2.75 (95% CI: 1.14-6.65; P = 0.03). TI occurring in clinical practice is associated with an increased risk of clinical progression; hence, it should be discouraged outside strictly experimental settings.