Different opinions exist about grading > 50% stenosis resulting from differences in angiographically estimated measurements and differences in weighing hemodynamic parameters at duplex scanning. The aim of this study was to reevaluate the correlation between duplex scanning measurements and angiographic findings in > 50% stenosis by using correlation coefficients instead of earlier determination coefficients in multiregression analysis in prediction of an absolute percentage of stenosis > or = 50%. The authors correlated the angiographic findings for 58 vessels with 50%-99% stenosis with findings at duplex examination. Peak systolic velocity (PSV) > 1.2 m/s was the limit for > 50% stenosis. The degree of stenosis was estimated as the smallest diameter of the diseased vessel divided by normal diameter of the vessel at the same level. Predictive values of different variables either alone or in combination were calculated by means of multiregression analysis. The highest predictive value was PSV, followed by late diastolic velocity (LDV) and pulse pressure (PSV-LDV) according to multiregression analysis. The results of Doppler evaluations of periorbital flow (POF) enhanced the differentiation between > 75% stenosis and < 75% stenosis. The combination of PSV and LDV and the result of POF coded as normal and abnormal had the highest accuracy in the prediction of absolute percentage of stenosis with a difference of 6.3 +/- 4.3% (SD) between the results of duplex scanning and angiography (r = 0.88). The accuracy in discriminating instances in which there was at least 75% reduction in luminal diameter was 98.2%. We found the quotient between PSV in the internal carotid artery and PSV in the common carotid artery to be less predictive with an accuracy of 84%. However, in the presence of severe stenosis in the external carotid artery, periorbital examination is not reliable. In such a situation it is preferable to use an equation based on pulse pressure and LDV. In prediction of > 50% stenosis we found multiregression analysis in the assessment of the predictive value of combined variables to be more accurate than single regression of each variable.