Using a methodology that has advanced the study of alcohol expectancies, the authors modeled a semantic network of marijuana expectancies stored in memory. They used individual-differences scaling, a form of multidimensional scaling, to map expectancies into memory network format and preference mapping to model likely paths of expectancy activation. Organization and activation of marijuana expectancies were found to vary with consumption level. Heavy marijuana consumers emphasized a relaxed-agitated dimension and were most likely to activate positive expectancies of relaxation, cognitive enhancement, and social facilitation. Nonconsumers, however, emphasized a detached-aware dimension and were more likely to activate negative expectancies of cognitive impairment and social impediment. Future efforts to alter likely activation patterns may be successful in changing use patterns.