We evaluated the association of the number of late antiretroviral therapy (ART) refills with patient outcomes in a large public-sector human immunodeficiency virus treatment program in Lusaka, Zambia. Using pharmacy data routinely collected during 2004-2010, we calculated the number of late refills during the initial year of ART. We used multivariable Cox proportional hazard regression to examine the association between the number of late refills and death or program failure (i.e., death, loss to follow-up, or program withdrawal) >12 months after ART initiation, with and without stratification by the medication possession ratio (MPR) during the initial year of ART. Of 53,015 adults who received ART for ≥12 months (median follow-up duration, 86.1 months; interquartile range, 53.2-128.2 months), 26,847 (50.6%) had 0 late refills, 16,762 (31.6%) had 1, 6,505 (12.3%) had 2, and 2,901 (5.5%) had ≥3. Kaplan-Meier analysis revealed that ≥3 late refills was associated with a greater mortality risk than 1 and 2 late refills (p<0.001, by the log-rank test). The mortality risk was greater for patients with 2 late refills [adjusted hazard ratio (HR), 1.17; 95% confidence interval (CI), 0.99-1.38] or ≥3 late refills (adjusted HR, 1.51; 95% CI, 1.23-1.87), compared with that for patients with 0-1 late refills. Program failure was associated with ≥2 late refills. An MPR of <80% was associated with similar increases in mortality risk across late-refill strata. Monitoring late refills during the initial period of ART may help resource- and time-constrained clinics identify patients at risk for program failure.