Specialization in violence is an important scientific and policy topic, and over the past several decades, many analysis techniques for studying specialization have emerged. Research in this area continues to be hampered, however, by remaining methodological problems. To overcome these problems, we propose a new method for studying specialization in violence based on an item-response theory measurement approach that is implemented through a multilevel regression model. Our approach defines specialization as an individual level latent variable, takes into account the inherent confounds between specialization and overall level of offending, and gauges specialization relative to the population base rates of each offense. Our method also enables researchers to 1) estimate the extent and statistical significance of specialization, 2) assess the stability of specialization over time, and 3) relate specialization to explanatory variables. Using data from three studies, we found substantial levels of specialization in violence, considerable stability in specialization over time, and several significant and relatively consistent relationships of specialization to explanatory variables such as gender, parental education, and risk-seeking.