Background: In psychiatry hard outcomes such as death are currently not available. ‘Response’ to treatment is therefore defined by cutoffs in terms of a minimum percentage change of a rating scale since the beginning of a trial. However, there is no consensus about the appropriate cutoff because a high variety has been used (e.g. at least 20%, 30%, 40% or 50% reduction) and the clinical meaning of the different cutoffs is unclear. This lack of a clinically meaningful standard is a problem for systematic reviews, because there is uncertainty about which criterion should be used. Objectives: To develop clinically meaningful response criteria by linking percentage reduction of total scores of complex rating scales with clinical global impressions in schizophrenia. Methods: Three large databases were available for analysis including the individual results from several thousands of people with schizophrenia in randomised antipsychotic drugs trials. The participants were simultaneously rated with comprehensive rating scales (Positive and Negative Syndrome Scale (PANSS, 30 items, total score 30 to 210) or Brief Psychiatric Rating Scale (BPRS, 18 items, score 18 to 126)), and the simple Clinical Global Impression Scale (CGI, 1 item score from 1 = very much better to 7 = very much worse). Percentage PANSS/ BPRS reduction from baseline was compared with the CGI improvement score using equipercentile linking, a method that uses the percentile rank functions of both variables for their comparison. Results: To be ‘minimally improved’ according to the CGI score was associated with a mean percentage reduction of the PANSS/BPRS of approximately 25%. To be ‘much improved’ was associated with approximately 50% PANSS/BPRS reduction from baseline. The linking functions were rather independent of participants’ severity at baseline. The results were found to be robust by replications in independent databases. We suggest that >= 50% PANSS/BPRS reduction should be the response criterion for acutely ill people with schizophrenia. Conclusions: Equipercentile linking can help to develop clinically meaningful criteria to define response to treatment. These criteria can now be applied as standards in clinical trials and Cochrane reviews. We are currently applying the method to develop similar criteria for depression and anxiety.