Purpose Offence paralleling behaviour (OPB) is a relatively new concept that emphasizes the assessment and modification of behavioural patterns that parallel criminal offending. Extant empirical research into OPB has typically focussed on the similarity between aggressive behaviours observed in custody and violent index offences perpetrated in the community; the results of these studies have been inconsistent, with the degree of similarity varying within subjects and across studies; (Daffern, Howells, Stacey, Hogue, & Mooney, 2008; Daffern, Howells, Mannion, & Tonkin, 2009). The aim of this study was to establish the level of similarity between OPB predictions and actual in‐patient aggressive behaviours. Method This prospective pilot study used a novel practice algorithm (Jones, 2010a) to develop a reference formulation from which OPBs and Pro‐social alternate behaviours (PABs), the pro‐social variants of OPBs, were predicted. Participants were five mentally disordered offenders resident in a UK medium secure psychiatric unit. Following generation of a reference formulation and OPB and PAB predictions the participants were monitored for 6 months. The degree of similarity between predicted and actual OPBs and PABs was calculated using Jaccard's coefficient. Results Results indicated considerable similarity between matched (pairing predictions and their corresponding actual behaviours) OPBs, which were also more similar than random pairs (pairing randomly selected predictions and aggressive behaviours). Conclusion This study revealed the existence of OPB in mentally disordered offenders and has provided the first test of a novel practice algorithm, revealing its potential to guide OPB formulations.