Background: Knowing the time cannabis was last used is important for determining impairment in accident investigations and clinical evaluations. Two models for predicting time of last cannabis use from single plasma cannabinoid concentrations-model I, using ⌬ 9 -tetrahydrocannabinol (THC), and model II, using the concentration ratio of 11-nor-9-carboxy-THC (THCCOOH) to THC-were developed and validated from controlled drug administration studies. Objectives of the current study were to extend the validation by use of a large number of plasma samples collected after administration of single and multiple doses of THC and to examine the effectiveness of the models at low plasma cannabinoid concentrations. Methods: Thirty-eight cannabis users each smoked a 2.64% THC cigarette in the morning, and 30 also smoked a second cigarette in the afternoon. Blood samples (n ؍ 717) were collected at intervals after smoking, and plasma THC and THCCOOH concentrations measured by gas chromatography-mass spectrometry. Predicted times of cannabis smoking, based on each model, were compared with actual smoking times. Results: The most accurate approach applied a combination of models I and II. For all 717 plasma samples, 99% of predicted times of last use were within the 95% confidence interval, 0.9% were overestimated, and none were underestimated. For 289 plasma samples collected after multiple doses, 97% were correct with no underestimates. All time estimates were correct for 76 plasma