The new method focuses on the variety of offenses committed by an individual rather than on their sequence. Research on transitions between subsequent offenses does not provide full information on the pattern of offending by an individual. Also, the sequential approach requires information on the chronology and timing of all offenses, which is unavailable for many otherwise useful datasets. The new method also relies on self-report measures of offending, since this provides a wealth of information beyond that contained in official crime records. Further, the new method defines specialization at the individual level so as to permit regression modeling. Methods that define the extent of specialization for groups of offenders limit the research on the correlates of specialization to bivariate relationships with simple categorical variables. In addition, the method isolates specialization from the overall tendency to offend, so as to take into account differences in the preciseness of information on specialization that stems from variations in offense rate. Finally, the new method separates specialization from offense base rates. This recognizes that specialization in offending should be measured by the contrast between an individual's concentration of offenses in certain categories and the overall rate of those offenses in the population. The new approach to studying specialization in violence is illustrated with three datasets, two of which provide measures of self-reported delinquency for the same sample at various ages. These data permit an assessment of the consistency of findings across locations and age, as well as assessment of the stability of specialization over time. 5 tables, 58 references, and supplementary data