The hyperdense middle cerebral artery sign is a CT predictor of the development of a large cerebral infarct. The limits of detectability were tested in a blinded, then unblinded analysis of CT scans from 25 acute stroke patients. In the initial blinded analysis, sign detection exhibited the following mean values: sensitivity, 78.5%; specificity, 93.4%; positive predictive value, 66%; negative predictive value, 96.4%; accuracy, 91.3%. Kappa statistics analysis indicated poor interobserver agreement (k = .38). Results of unblinded analysis were as follows: sensitivity, 69%; specificity, 94.4%; positive predictive value, 82.8%; negative predictive value, 88.7%; accuracy, 87.3%. There was fair unblinded interobserver agreement (k = .53). Unblinded analysis had a lower false-positive frequency and did not increase the number of true-positive determinations. We conclude that detection of the hyperdense middle cerebral artery sign on CT scans by multiple observers is a sensitive, accurate, and predictive indicator of middle cerebral artery thromboembolism.