Full 12-hour pharmacokinetic profiles of nevirapine, stavudine, and lamivudine in HIV-infected children taking fixed-dose combination antiretroviral tablets have been reported previously by us. Further studies with these formulations could benefit from less-intensive pharmacokinetic sampling. Data from 65 African children were used to relate area under the plasma concentration versus time curve over 12 hours (AUC) to plasma concentrations of nevirapine, stavudine, or lamivudine at times t = 0, 1, 2, 4, 6, 8, and 12 hours after intake using linear regression. Limited sampling models were developed using leave-one-out crossvalidation. The predictive performance of each model was evaluated using the mean relative prediction error (mpe%) as an indicator of bias and the root mean squared relative prediction error (rmse%) as a measure of precision. A priori set criteria to accept a limited sampling model were: 95% confidence limit of the mpe% should include 0, rmse% less than 10%, a high correlation coefficient, and as few (convenient) samples as possible. Using only one sample did not lead to acceptable AUC predictions for stavudine or lamivudine, although the 6-hour sample was acceptable for nevirapine (mpe%: -0.8%, 95% confidence interval: -2.2 to +0.6); rmse%: 5.8%; r: 0.98). Using two samples, AUC predictions for stavudine and lamivudine improved considerably but did not meet the predefined acceptance criteria. Using three samples (1, 2, 6 hours), an accurate and precise limited sampling model for stavudine AUC (mpe%: -0.6%, 95% confidence interval: -2.2 to +1.0; rmse%: 6.5%; r: 0.98) and lamivudine AUC (mpe%: -0.3%, 95% confidence interval: -1.7 to +1.1; rmse%: 5.6%; r: 0.99) was found; this model was also highly accurate and precise for nevirapine AUC (mpe%: -0.2%, 95% confidence interval: -1.0 to +0.7; rmse%: 3.4%; r: 0.99). A limited sampling model using three time points (1, 2, 6 hours) can be used to predict nevirapine, stavudine, and lamivudine AUC accurately and precisely in HIV-infected African children.