In this paper, a new method is proposed for reconstructing the three-dimensional (3D) structure of cerebral arteries from the digital subtraction angiographic (DSA) image sequence, which is taken by the rotational stereo angiographic system. The DSA image sequence is arranged in order of rotation angle to constitute the spatio-temproral image. The sinusoidal movement locus of each vessel can be observed on the transversal sections of spatio-temproral image. We prove that the 3D coordinate of each vessel can be estimated from the parameters of sinusoidal wave form of the movement locus. Such parameters, amplitude and phase in sinusoidal wave form, are obtained by utilizing the voting algorithm anologous to the Hough transform. In this method, an accurate parameter estimation can be performed even from the limited data. The 3D structure of arteries is reconstructed by the 3D coordinate of each vessels. The validity of proposed reconstruction algorithm is confirmed by reprojecting the reconstructed arteries and comparing to the original DSA images. The structure of arterial tree in the reprojected image is closely resembled to that in the corresponding original DSA image. The reconstructed cerebral arteries are displayed by using the computer graphic software, and the 3D structure can be clearly observed from arbitrary view point.