The differentiation of cerebral infarction and cerebral haemorrhage is the most important first step in the management of acute stroke, because clinical management of the two disorders differs substantially. The Guy's Hospital and Siriraj stroke diagnostic scores were developed to aid clinicians in this decision. We have tested the performance of the two scores on a group of 1059 patients admitted to the acute stroke unit (ASU) at the Western Infirmary, Glasgow, with suspected stroke between May, 1990, and December, 1993. The diagnosis was confirmed as stroke by computed tomography (CT) scanning or necropsy (n = 10) in 991 patients. For each clinical score we subjectively identified the cut-off point that maximised sensitivity to cerebral haemorrhage with the smallest loss of specificity. At its optimum cut-off point the Guy's Hospital score had a sensitivity for the diagnosis of haemorrhage of 70% and specificity of 64%. The corresponding figures for the Siriraj score at its optimum cut-off point were 68% sensitivity and 64% specificity. The overall predictive accuracy of both scores was 64%. The greater complexity of the Guy's Hospital score (thirteen variables included) did not result in substantially superior performance to the Siriraj score (five variables). This validation study suggests that neither score is useful for exclusion of haemorrhage before anticoagulant treatment is initiated or as a diagnostic screening procedure for trials of low-risk treatments such as aspirin. Our findings emphasise the need for routine CT scanning in this setting, since this method remains the most accurate for differentiating between haemorrhage and infarction.