As identifying patients at risk of subsequent suicidal behaviour is a key goal of assessment, a cohort study of presentations to five emergency departments following episodes of self-harm was carried out. We compared the accuracy of the prediction of subsequent self-harm within 6 months between global clinical assessments and the Manchester Self-harm Rule. Sensitivity, specificity, and positive and negative predictive values with 95% confidence intervals (CI) were calculated. Global clinical assessments and the rule had a sensitivity of 85% (CI 83 to 87) versus 94% (CI 92% to 95%), specificity of 38% (CI 37% to 39%) versus 26% (CI 24% to 27%), a positive predictive value of 22% (CI 21% to 23%) versus 21% (CI 19% to 21%) and a negative predictive value of 92% (CI 91% to 93%) versus 96% (CI 94% to 96%). The accuracy of predicting short-term repetition of self-harm by clinicians could be improved by incorporating this simple rule into their assessment.