The variability of the pharmacokinetics of zidovudine after its oral administration to 36 AIDS patients has been investigated by measuring the plasma and urine levels of zidovudine and its metabolite on Days 1 and 35 of continuous treatment. A two-phase absorption model was first defined from well-documented data in 12 subjects. The population characteristics of the kinetic parameters for both days were estimated by a nonparametric method. On Day 1, the mean (coefficient of variation) volume of distribution of zidovudine was 94.41 (90%), its mean half-life was 0.81 h (107%) and its mean oral clearance was 117 l.h-1 (57%) and on Day 35, these values were, respectively, 1121 (139%), 0.75 h (181%) and 295 l.h-1 (196%). The results confirm the large interindividual and intraindividual variation in zidovudine kinetics. The four covariates included in the population analysis (body weight, serum haemoglobin, creatinine and bilirubin) did not show clear relationship to the kinetic parameters. Thirty-four subjects were follow-up clinically for 99 days to 367 days after initiation of zidovudine therapy. The relationship between individual kinetic parameters (determined by Bayesian estimation), mean concentration profiles and outcome was studied through survival analysis. Long-term efficacy was defined as the prevention of opportunistic infections, which occurred in 13 patients. No clinical or kinetic variables, nor the individual zidovudine concentration profiles were found to predict the occurrence of an opportunistic infection. Toxicity was defined as a 20%-decrease in serum haemoglobin, which occurred in 13 patients. A significant relationship between mean daily concentration and toxicity was found, with an hazard of occurrence of toxicity 4.3-times larger when the mean steady stade concentration was 0.8 mg.l-1 than 0.6. The results indicate that zidovudine dosage should be individualised.