We predicted that the closer the resemblance of attempted suicide to completed suicide, the greater the risk of patients later killing themselves after an unsuccessful attempt. Using two risk scales, one of 6 and the other of 18 socio-demographic and clinical items, weighted according to their value in discriminating between suicides and attempted suicides, and a rating of intent to die during the index suicide attempt, we followed up 1263 attempted suicides for two years. Ten of the twelve suicidal deaths in the first year occurred in patients scoring in the top quartile on the 6-item scale; and among high-risk scorers there was an excess of those defined as 'failed suicides' at the index attempt. The 18-item scale was superior to the shorter one, but both scales were improved by taking into account the rating of intent-to-die. The potential value of these findings is discussed in the light of some inherent problems of predicting suicide in the clinical setting.