OBJECTIVE: To determine the relationship between mortality risk and the CD4 cell response to antiretroviral therapy (ART). DESIGN: Observational community-based ART cohort in South Africa. METHODS: CD4 cell counts were measured 4 monthly, and deaths were prospectively ascertained. Cumulative person-time accrued within a range of updated CD4 cell count strata (CD4 cell-strata) was calculated and used to derive CD4 cell-stratified mortality rates. RESULTS: Patients (2423) (median baseline CD4 cell count of 105 cells/microl) were observed for up to 5 years of ART. One hundred and ninety-seven patients died during 3155 person-years of observation. In microltivariate analysis, mortality rate ratios associated with 0-49, 50-99, 100-199, 200-299, 300-399, 400-499 and at least 500 cells/microl updated CD4 cell-strata were 11.6, 4.9, 2.6, 1.7, 1.5, 1.4 and 1.0, respectively. Analysis of CD4 cell count recovery permitted calculations of person-time accrued within these CD4 cell-strata. Despite rapid immune recovery, high mortality in the first year of ART was related to the large proportion of person-time accrued within CD4 cell-strata less than 200 cells/microl. Moreover, patients with baseline CD4 cell counts less than 100 cells/microl had much higher cumulative mortality estimates at 1 and 4 years (11.6 and 16.7%) compared with those of patients with baseline counts of at least 100 cells/microl (5.2 and 9.5%) largely because of greater cumulative person-time at CD4 cell counts less than 200 cells/microl. CONCLUSION: Updated CD4 cell counts are the variable most strongly associated with mortality risk during ART. High cumicrolative mortality risk is associated with person-time accrued at low CD4 cell counts. National HIV programmes in resource-limited settings should be designed to minimize the time patients spend with CD4 cell counts less than 200 cells/microl both before and during ART.