This study aimed to examine the number of latent classes of criminal social identity that exist among male recidivistic prisoners. Latent class analysis was used to identify homogeneous groups of criminal social identity. Multinomial logistic regression was used to interpret the nature of the latent classes, or groups, by estimating the associationsto number of police arrests, recidivism, and violent offending while controlling for current age. The best fitting latent class model was a five-class solution: 'High criminal social identity' (17%), 'High Centrality, Moderate Affect, Low Ties' (21.7%), 'Low Centrality, Moderate Affect, High Ties' (13.3%),'Low Cognitive, High Affect, Low Ties' (24.6%), and 'Low criminal social identity' (23.4%). Each of the latent classes was predicted by differing external variables. Criminal social identity is best explained by five homogenous classes that display qualitative and quantitative differences.