Two mathematical models are described for the prediction of time of marijuana use from the analysis of a single plasma sample for cannabinoids. The models were derived from cannabinoid data obtained from a controlled clinical study of acute marijuana smoking. Model I was based on plasma delta 9-tetrahydrocannabinol (THC) concentrations and Model II was based on the ratio of 11-nor-9-carboxy-delta 9-tetrahydrocannabinol (THCCOOH) to THC in plasma. The two models were validated with cannabinoid data from nine published and unpublished clinical studies. The data included plasma samples obtained from infrequent and frequent marijuana smokers and after oral marijuana administration. Cannabinoid plasma concentrations had been determined by a variety of analytical methods. The accuracy of model prediction was evaluated by comparison of the predicted time of prior drug use to the actual time of exposure. Predictions of time of exposure were generally accurate but tended to overestimate time immediately after smoking and tended to underestimate later times. A second assessment of the validity of the models was made by determining if actual time of use was within the 95% confidence interval. Model I correctly predicted the time of exposure within the 95% confidence interval for 235 of 261 samples (90.0%), and Model II was correct in 232 of 260 samples (89.2%). These prediction models may be beneficial to forensic scientists in the interpretation of cannabinoid plasma levels.