Deconvolution methods are not feasible in spectral domain optical coherence tomography (SD-OCT) system due to the linear space variant property of its axial point spread function (PSF). In order to improve the axial resolution and enhance the imaging quality of the SD-OCT system, a Lucy-Richardson algorithm based deconvolution method is implemented with the numerically compensated A-scan signals and the effective axial PSF in this paper. The main factors decreasing the value of PSF along the axial direction and blurring the images, as well as the deconvolution process, are theoretical studied. With the developed SD-OCT system, the axial modulation function of the PSF is retrieved by curve fitting of the measured peak values of the PSF at different axial positions. After compensating the A-scan signal with the modulation function, deconvolution is performed with the measured effective axial PSF. In vivo imaging on typical samples is conducted and reconstruction is done based on the deconvolution method. The comparing results demonstrate that the proposed deconvolution method not only enhances the axial resolution but also improves the depth dependent sensitivity fall-off in SD-OCT system.