We employ two-stage least squares to estimate causal effects in structural equation models specific to six different crimes and two noncriminal deviant acts using data from a large-scale survey of adults. The findings indicate: (1) As Sutherland claimed, excess association with definitions favorable to crime and/or deviance tends to increase crime/deviance (measured here by self-predictions), but this effect is mainly indirect, through increasing motivation to engage in deviant acts; (2) the same differential association process holds for a range of crimes and also for the two noncriminal deviant acts, suggesting that Sutherland's theory is more widely applicable than he himself claimed. Our analysis also suggests one conclusion that is more speculative: Definitions (social messages) favorable to different forms of crime/deviance may form several independent domains rather than a single general domain referring to all crime/deviance. (Author abstract)