Abstract: In the 3D measurement of photometric stereo, 3D surface depth is reconstructed from 2D gradient fields by reconstruction algorithm which takes lots of time in high resolution condition. In this paper, a fast reconstruction algorithm of photometric stereo was proposed by using the improved multi-scale Haar wavelet transform. The improvement was to replace the traditional wavelet channels with the gradient fields in wavelet transformation, which was applied in 3D surface reconstruction of photometric stereo to reduce time consumption. The experiments showed that time consumption of the proposed algorithm was linear to the resolution of gradient fields, while the traditional algorithms were higher than linear. Besides, more robustness and less error appeared in the results of proposed algorithm.