Background: The ABCD2 stroke risk score is recommended in national guidelines for stratifying care in transient ischaemic attack (TIA) patients, based on its prediction of early stroke risk. We had become concerned about the score accuracy and its clinical value in modern TIA cohorts. Methods: We identified emergency department-diagnosed TIA at two hospitals over 3 years (2004-2006). Cases were followed for stroke occurrence and ABCD2 scores were determined from expert record review. Sensitivity, specificity and positive predictive values (PPV) of moderate-high ABCD2 scores were determined. Results: There were 827 indexed TIA diagnoses and record review was possible in 95.4%. Admitted patients had lower 30-day stroke risk (n= 0) than discharged patients (n= 7; 3.1%) (P < 0.0001). There was no significant difference in proportion of strokes between those with a low or moderate-high ABCD2 score at 30 (1.2 vs 0.8%), 90 (2.0 vs 1.9%) and 365 days (2.4 vs 2.4%) respectively. At 30 days the sensitivity, specificity and PPV of a moderate-high score were 57% (95% confidence interval (CI) 25.0-84.2), 32.2% (95% CI 29.1-35.6) and 0.75% (95% CI 0.29-1.91) respectively. Conclusions: Early stroke risk was low after an emergency diagnosis of TIA and significantly lower in admitted patients. Moderate-high ABCD2 scores did not predict early stroke risk. We suggest local validation of ABCD2 before its clinical use and a review of its place in national guidelines.