Research on offense specialization concludes that there is a great deal of versatility in offending. After research designs predetermined offense versatility, the consensus became that general offending trumps offense specialization. However, the method, marginal logit modeling, preserves a nominal measurement of the breadth of criminal offending and allows respondents to contribute as many affirmative responses as necessary to fully represent their offense history. Marginal logit modeling is proposed as an alternative method that supports many desirable features suited to the investigation of offense specialization. This paper assesses offense specialization by using the marginal logit model which captures and quantifies the correlation between criminal behaviors, conditional on the independent variables. The analysis puts offense specialization on an equal footing with generalized offending. The paper analyzes nine self-reported delinquent behaviors. It shows that violent offenders are more likely to engage in additional violent offenses and nonviolent offenders are more likely to engage in additional nonviolent offenses. The results suggest that some special causal processes are involved in specific offenses and caution needs to be taken when making sweeping generalizations about the effects on crime or violent crime. Tables and references