This study evaluated the discriminative stimulus effects of caffeine in humans. Nine normal male and female volunteers (ages 18-28) were trained to discriminate between the methylxanthine central nervous system stimulant caffeine (320 mg/70 kg, p.o.) and placebo. Monetary reinforcers were earned for identifying correctly the letter code associated with each substance. After four training sessions, the ability to discriminate between the two training conditions was tested for 20 sessions (test-of-acquisition). In subjects who met the criterion for discrimination (i.e., greater than or equal to 85% correct responding on greater than or equal to 4 consecutive sessions during the last 10 test-of-acquisition sessions) dose-effect curves for caffeine (0, 56, 100, 180, 240 and 320 mg/70 kg) and for the benzodiazepine triazolam (0, 0.10, 0.32 and 0.56 mg/70 kg) were determined. Seven of the nine subjects learned the caffeine-placebo discrimination and their performance improved across the 20 test-of-acquisition sessions. The training dose of caffeine (320 mg/70 kg) produced stimulant-like self-reports that differed from placebo when the letter codes were identified correctly, but not when the letter codes were identified incorrectly. Novel caffeine doses produced dose-related increases in caffeine-appropriate responding (N = 4), whereas triazolam produced predominantly placebo-appropriate responding (N = 3) and self-reports that differed from both caffeine and placebo. Throughout dose-effect curve determinations, the training dose of caffeine and placebo continued to be identified correctly (range: 84-100% correct responding). These results suggest that a caffeine (320 mg/70 kg)-placebo discrimination 1) can be acquired and maintained, 2) is related to self-reported drug effects if the training conditions are correctly identified and 3) has some pharmacological specificity.