In order to reduce the error produced by using the white light interferometry while calibrating the microcosmic shape, an improved Carré phase shifting algorithm based on white light interferometry was proposed. The search path of traditional extremum algorithm was optimized, and then combined with the Carré phase shifting algorithm to extract the phase. The proposed algorithm can eliminate the random error caused by the shot-noise of CCD, it do not need to limit the step of the phase adjustments by using the central wavelength of white light and the phase unwrapping is no longer needed, and it makes the computation efficiency improved. The central wavelength on line was modified to reduce the error caused by the characteristic of light and environment disturbance. The three-dimensional topography of standard roughness specimens was restored and the surface roughness was mesuresed by different kinds of algorithms. The experimental results show that, compared with other algorithms, the accuracy is improved and repeatability is better than 1% by using the proposed improved algorithm.