Unpronounceable strings of 4 consonants (conditioned stimuli: CSs) were consistently followed by familiar words belonging to 1 of 2 opposed semantic categories (unconditioned stimuli: USs). Conditioning, in the form of greater accuracy in rapidly classifying USs into their categories, was found when visually imperceptible (to most subjects) CSs occupied ≥58 ms of a 75-ms CS–US interval. When clearly visible CSs were presented in a 375 ms CS–US interval, conditioning was strongly correlated with measures of contingency awareness, and did not occur in the absence of that awareness. These experiments delineated 2 forms of conditioning: Unconscious conditioning occurred with a brief CS–US interval, with an effectively masked conditioned stimulus (CS), and with no reportable knowledge of the contingent CS–US relation. Conscious conditioning occurred with a substantially longer CS–US interval, a perceptible CS, and with subjects’ reportable knowledge of the contingent CS–US relation.