A windowed Fourier transform (WFT) based method is proposed for extracting and compensating depth-resolved phase error in spectral domain optical coherence tomography (SDOCT) system. Firstly, by using a WFT to the interference spectrum of the light from the sample and the reference mirror, the depth-frequency distribution of A-scan of the sample is obtained. Due to the time-frequency characteristics of the WFT, the interference spectra corresponding to different interfaces at different depths are separated. The polynomial fitting for the phase variation of each complex interference spectrum is then performed and the phase errors distribution with the change of depth is obtained. Based on these phase errors, a precise numerical compensation for the phase is carried out. This method can not only be applied for extracting and compensating of depth-resolved dispersion phase error, but also can be used for depth-varied phase error compensation resulted from uneven spectrum sampling in wave-number space. A simulation for dispersion phase error extraction is conducted. Finally, the SDOCT images of the 4-layer cover glasses and fingernail of a volunteer are obtained and then used for phase error extraction with the WFT method. The results demonstrate that the proposed method has the capability of extracting the phase errors with high precision, leading to the improvement of the depth resolution and the image quality after phase error compensation.