Background In schizophrenia, the presence of certain cognitive biases has been established. Informed by this, metacognitive training (MCT) has been developed for schizophrenia. There is increasing evidence of its effectiveness with some patients, but its applicability to dangerous patients has not yet been demonstrated. Aims Our aim was to test the feasibility of a randomised controlled trial (RCT) of MCT for patients in a high‐security hospital setting. Methods Twenty of 33 eligible and selected male in‐patients with schizophrenia and a history of violence were randomised pairwise to eight sessions of MCT or treatment as usual. Symptom severity and reasoning, according to the jumping to conclusions paradigm, were measured before, immediately after treatment, and 3 and 6 months later. Results Men in both groups completed the trial, and those in the MCT arm, almost all of the group sessions. The MCT arm had a significant advantage in improvement of ‘suspiciousness’, greatest at 3 months, but then declining. No significant improvement in reasoning ability was achieved. Conclusions Metacognitive training showed sufficient promise in this group for a full trial to be worthwhile, and the feasibility of an RCT methodology, even in a secure hospital, was established. The fact that the improvements faded during follow‐up suggests that a useful modification to the treatment would be lengthening the protocol, repeating it, or both. Copyright © 2014 John Wiley & Sons, Ltd.