The current study aimed to develop and validate the Measure of Delinquent Social Identity (MDSI). Dimensionality and construct validity of the MDSI was investigated in a sample of youth offenders (N = 536). Four alternative models of the MDSI were estimated using Mplus. The model identified as being the best fit for the data was a bifactor model with three dimensions (cognitive centrality, in-group affect, in-group ties), while controlling for the general factor. The three subscales differentially correlated with criminal friend index, self-esteem, parental attachment and peer rejection. Limitations and advantages, including practical implications, of the current research are discussed.