Indirect memory associations for substance use predict both the concurrent and prospective levels of substance use. These methods assess spontaneous, possibly implicit, and easily accessible associations that predict substance use over direct (explicit) methods of assessment (e.g., outcome expectancies). The present study tested and expanded the application of a coding method for alcohol and marijuana associations on the basis of self-coding of indirect responses (Frigon & Krank, 2009). College students generated free associates to (1) ambiguous words (e.g., draft or weed), (2) situations (e.g., at a party, hanging out with friends), and (3) emotions (having fun, feeling dreamy). Later, participants were shown their responses and were asked to code their responses according to both nonrisk and risk activities, such as alcohol and marijuana use. Self-coded scores were higher than researcher-coded scores, captured the same variance, and improved the prediction of substance use. Self-coding of indirect memory associations provides accurate and efficient prediction of the level of alcohol and marijuana. Self-coding is efficient and may be useful for reducing ambiguities in coding of many different kinds of open-ended responses.