The prognosis of individual patients with transient ischaemic attacks (TIAs) is extremely variable; some patients are at high risk and others at low risk of a serious vascular event. Prediction equations of outcome were developed, based on eight clinical prognostic factors, from a cohort of 469 hospital-referred TIA patients ("training" data set), that enable high (and low) risk patients to be identified and for whom costly and risky treatments may (or may not) be targeted. The study aimed to determine whether these equations are externally valid and can predict outcome, with reliability and discrimination, in two independent cohorts of TIA patients ("test" data sets): 1653 TIA patients in the UK-TIA aspirin trial and 107 TIA patients in the Oxfordshire Community Stroke Project. Predicted outcomes agreed closely with the observed outcomes in the "test" data sets (reliability) for all outcome events at low five year risk (< 30%) but the estimates of risk were less precise in groups predicted to have a high five year risk (> 40%). The prediction equations were fairly accurate in discriminating between patients who subsequently suffered the outcome event of interest and those who survived free of the event at five years after the TIA, particularly at lower cut-off levels distinguishing high and low risk (for example, < 30% vs > 30% at five years). It is very difficult to achieve perfect discrimination because there is no single important prognostic factor for TIA patients that indicates whether a patient is going to suffer an event or not. These equations can be used to provide a reliable estimate of the absolute five year risk of a serious vascular event in hospital-referred TIA patients but they cannot, as yet, be used with confidence to distinguish patients at high risk from patients at low risk.