To try and improve the interobserver agreement for the diagnosis of TIA, we used a checklist in which the symptoms were recorded in plain language, instead of in abstract diagnostic terms such as amaurosis fugax. Criteria for a diagnosis of TIA were similarly phrased and recommended to all observers. Eight senior neurologists and ten neurology residents interviewed 72 patients in random pairs. In 64 cases the observers agreed on the diagnosis (kappa value = 0.77). After a short discussion between the two observers the agreement increased to a maximum (kappa = 1.0). However, in 29 of the 144 interviews the diagnosis would have been different, had the recommended criteria been fully applied. In 28 of these the observer had diagnosed TIA on insufficient evidence. Six "misinterpretations" led to disagreement for the diagnosis and only these were corrected by the observers during their discussion. The present design has led to a maximal agreement for the diagnosis of TIA between two observers, but the agreement between such a pair and the common diagnostic criteria was not yet ideal. The precision of the diagnosis could be improved if details of the required symptoms are discussed in general as well as for each patient.