Although the psychological treatment of addiction is premised on the notion that addictive behavior is learned behavior, addiction workers have been slow to exploit the capacity of the computer to promote new learning. The best computer-assisted learning programs are fully adaptive; their design anticipates the responses a user can make, classifies them, and provides feedback for each. This paper reports on two recent attempts to achieve this level of sophistication in computer-assisted drug prevention. One program seeks to prevent heroin relapse among currently abstinent prisoners, and the other promotes controlled drinking in alcohol abusers. Given the fact that most drug users can be expected to opt for self-help materials over the offer of formal therapy, and that most (licit) drug users who solve their addiction problems do so without recourse to professional help anyway, the use of computer-assisted drug prevention programs like these provides an important new direction in substance abuse treatment.