Psychological treatments targeting emotion dysregulation in adolescents are effective in reducing nonsuicidal self-injury (NSSI) but predicting treatment outcome remains difficult. Identifying sub-groups based on repeated measurements of emotion dysregulation pre-treatment may guide personalized treatment recommendations. We used data from a recent trial evaluating internet-delivered emotion regulation therapy for adolescents with NSSI (n = 138). Latent class analysis was used to identify sub-groups based on pre-treatment responses on the 16-item version of the Difficulties in Emotion Regulation Scale. The primary outcome was self-rated NSSI frequency during treatment, and secondary outcome was the proportion of participants with no NSSI 1-month post-treatment. Three sub-groups of emotion dysregulation were identified: Low variability and low mean (Group 1), Low variability and high mean (Group 2), and High variability and low mean (Group 3). Sub-groups did not differ in NSSI frequency during treatment (Group 2 IRR = 1.06 [95% CI 0.49 – 2.29], p = .88; Group 3 IRR = 1.22 [95% CI 0.31 – 4.76], p = .77). However, more participants in Group 1 compared to Group 2 abstained from NSSI at 1-month post-treatment (OR = 3.63 [95% CI 1.16 – 11.33], p = 0.01). Latent class analysis identified sub-groups predictive of NSSI absence post-treatment, demonstrating clinical utility.