A limited sampling strategy was developed to estimate the total area under the curve of plasma cyclophosphamide concentrations versus time (AUC). The strategy was developed with a training set consisting of 29 pharmacokinetic studies in 16 patients who received 1-h i.v. infusions of cyclophosphamide at a dosage of 1000 mg/m2. The strategy was developed by applying stepwise forward multiple regression analysis to cyclophosphamide concentrations observed at each time in the training set (independent variables) versus the AUC (dependent variable). It was confirmed by applying stepwise backward elimination regression analysis to the same data set. The final sampling strategy, which utilized three time points, was: AUC = 40.18C24 + 8.79C4 + 0.83C1 - 28 (dosage/1000), where C24, C4, and C1 represent plasma cyclophosphamide concentrations at 24, 4, and 1 h, respectively, and the dosage is in mg/m2 (r = 0.98). The strategy was validated prospectively with a test data set consisting of 14 pharmacokinetic studies in 11 patients who received 1-h i.v. infusions of cyclophosphamide at dosages of 300, 600, or 1200 mg/m2. The strategy proved highly predictive, with correlation coefficient between predicted and actual AUC of 0.94. The strategy also proved unbiased, with mean percentage of error (+/- SE) of 3.3 +/- 3.6%, and precise, with mean absolute percentage of error of 9.3 +/- 2.7%. The sampling strategy developed is being used in a multiinstitution trial of cyclophosphamide in an effort to relate cyclophosphamide pharmacokinetics, as expressed by AUC, with the toxic or therapeutic pharmacodynamic responses of the drug.