The traditional vibration measurement method usually uses contact sensors, which induce unwanted mass and have limits for moving parts. Currently, using stereo vision for vibration measurement is developing fast because of noncontact and full field. However, for stereo vision, the high-accuracy reconstruction for the vibration motion is a significant challenge and the factors that affect the construction accuracy have not been thoroughly studied. The accuracy analysis of sinusoidal motion reconstruction is important because it can provide guidance for the reconstruction of other vibration motions. The high accuracy reconstruction for sinusoidal motion and the factors that affect the accuracy of the reconstruction are presented. First, the error model of reconstruction using stereo vision considering the delay time, frequency, amplitude, and disparity is established. The accuracy of sinusoidal motion reconstruction in the whole period is analyzed theoretically and experimentally considering subpixel interpolation. Peak identification is essential for the sinusoidal motion reconstruction and has the highest resolution requirement for the reconstruction resolution. This requirement is systematically investigated for the sinusoidal motions with different frequencies and amplitudes. The relationships between the reconstruction resolution and parameters of cameras are analyzed.