There are several clinical scenarios in which knowledge of zidovudine disposition may be important. This study evaluated the clinical utility of pharmacokinetic parameters for zidovudine derived from sparse serum concentration data obtained in an outpatient setting. Twelve human immunodeficiency virus-infected participants had two serum zidovudine concentration determinations obtained on two different clinic visits, 2 to 38 days apart. Zidovudine concentrations were measured by radioimmunoassay. A one-compartment oral absorption model was used to describe zidovudine disposition. Three different approaches were used to estimate pharmacokinetic parameters: Bayesian estimation with one or two concentrations and least squares with one concentration. The ability of these parameters to predict concentrations measured during the second clinic visit was assessed by calculation of precision and bias and compared with predictions using standard fixed or weight-adjusted parameters. Estimated pharmacokinetic parameters for zidovudine were consistent with literature values; there was no statistically significant difference among the parameters calculated with the three estimation strategies. Absorptive phase concentrations were poorly predicted by all methods (mean percent bias, 157 to 249%; mean percent precision, 389 to 537%). Predictive ability for concentrations obtained in the elimination phase was strikingly improved: mean percent bias, -17 to 70%; mean percent precision, 40 to 95%. Bayesian and least-squares estimated parameters were statistically better than fixed-parameter values for predicting concentrations in the elimination phase. These observations provide a modeling framework to determine pharmacokinetic disposition of zidovudine in an individual, screen for the existence of a drug interaction, and conduct concentration-controlled clinical trials.