BACKGROUND: General practitioners are usually the first health professionals to be contacted by people with early signs of psychosis. We aimed to assess whether increased liaison between primary and secondary care improves the clinical effectiveness and cost-effectiveness of detection of people with, or at high risk of developing, a first psychotic illness. METHODS: Our Liaison and Education in General Practices (LEGs) study was a cluster-randomised controlled trial of primary care practices (clusters) in Cambridgeshire and Peterborough, UK. Consenting practices were randomly allocated (1:1) to a 2 year low-intensity intervention (a postal campaign, consisting of biannual guidelines to help identify and refer individuals with early signs of psychosis) or a high-intensity intervention, which additionally included a specialist mental health professional who liaised with every practice and a theory-based educational package. Practices were not masked to group allocation. Practices that did not consent to be randomly assigned comprised a practice-as-usual (PAU) group. The primary outcome was number of referrals of patients at high risk of developing psychosis to the early intervention service per practice site. New referrals were assessed clinically and stratified into those who met criteria for high risk or first-episode psychotic illness (FEP; together: psychosis true positives), and those who did not fulfil such criteria for psychosis (false positives). Referrals from PAU practices were also analysed. We assessed cost-effectiveness with decision analytic modelling in terms of the incremental cost per additional true positive identified. The trial is registered at the ISRCTN registry, number ISRCTN70185866. FINDINGS: Between Dec 22, 2009, and Sept 7, 2010, 54 of 104 eligible practices provided consent and between Feb 16, 2010, and Feb 11, 2011, these practices were randomly allocated to interventions (28 to low intensity and 26 to high intensity); the remaining 50 practices comprised the PAU group. Two high-intensity practices were excluded from the analysis. In the 2 year intervention period, high-intensity practices referred more FEP cases than did low-intensity practices (mean 1.25 [SD 1.2] for high intensity vs 0.7 [0.9] for low intensity; incidence rate ratio [IRR] 1.9, 95% CI 1.05-3.4, p=0.04), although the difference was not statistically significant for individuals at high risk of psychosis (0.9 [1.0] vs 0.5 [1.0]; 2.2, 0.9-5.1, p=0.08). For high risk and FEP combined, high-intensity practices referred both more true-positive (2.2 [1.7] vs 1.1 [1.7]; 2.0, 1.1-3.6, p=0.02) and false-positive (2.3 [2.4] vs 0.9 [1.2]; 2.6, 1.3-5.0, p=0.005) cases. Referral patterns did not differ between low-intensity and PAU practices. Total cost per true-positive referral in the 2 year follow-up was £26,785 in high-intensity practices, £27,840 in low-intensity practices, and £30,007 in PAU practices. INTERPRETATION: This intensive intervention to improve liaison between primary and secondary care for people with early signs of psychosis was clinically and cost effective. FUNDING: UK National Institute for Health Research.