As members of the recent National Institute for Health and Care Excellence (NICE) clinical guideline update for schizophrenia 1 (M.B. and D.S.), we read with interest the excellent meta-analysis of CBT for symptoms of schizophrenia by Jauhar et al. Reference Jauhar, McKenna, Radua, Fung, Salvador and Laws2 The results are broadly in line with the NICE review and particularly that of Wykes et al, Reference Wykes, Steel, Everitt and Tarrier3 which showed that studies with high methodological rigour, including masking, have a small effect size for positive and total symptoms. Clearly cognitive-behavioural therapy (CBT) is no panacea; but neither is it ineffective. Meta-analyses bring together all trials, with patients drawn from heterogeneous populations, including different phases of illness. The tests for heterogeneity not accounted for by chance (I Reference Jauhar, McKenna, Radua, Fung, Salvador and Laws2 ) in this meta-analysis were all high. The question therefore arises, ‘For whom is CBT in psychosis most effective and for what outcome?’ Likely groups are individuals at ultra-high risk for psychosis, 1 those in the early phase of psychosis 1 and perhaps those with chronic stable symptoms, appearing to benefit the most; the Prevention of Relapse in Psychosis trial suggests that those beginning treatment early in the course of recovery from acute symptoms do not benefit. These trials focus on individuals in receipt of medication, with enduring symptoms. They therefore ask the question, ‘Does CBT offer added value compared with medication alone?’ We might also ask the converse, ‘Does antipsychotic medication offer added value to CBT alone?’ It is known that up to 50% of individuals will not adhere to medication; a recent pilot trial of CBT in those not taking medication showed an effect of CBT equivalent to that of drugs. Reference Morrison, Turkington, Pyle, Spencer, Brabban and Dunn4 Given the low acceptability of antipsychotic medications and their serious impact on health, this is an important question for further research. 1 We note that our trial of CBT for commanding hallucinations is included in the analysis for hallucinations; however, this trial did not predict a reduction in hallucinations, but reported a ‘high’ effect size for harmful compliance (not reported), which has been the subject of a large multicentre trial, soon to report. We argued some time ago that CBT for psychosis should not be conceived and evaluated as a ‘quasi-neuroleptic’: Reference Birchwood and Trower5 the dimensions of delusions (power, distress) and general affective dysfunction are, we believe, among the most appropriate targets for CBT, with strong theoretical justification. Given the evidence from systematic reviews of antipsychotics Reference Lepping, Sambhi, Whittington, Lane and Poole6 that the improvements claimed for antipsychotics are of questionable clinical utility, with most trials failing to demonstrate minimal clinical improvement using the Positive and Negative Syndrome Scale, with effect sizes smaller than for adverse side-effects, there is clearly much work to be done to improve care, as the Schizophrenia Commission outlined in their 2012 review of current treatment and services (www.schizophreniacommission.org.uk).