We apply the multi-scale theory of wavelet transform to determine the optimal step parameter. More specifically, we decompose the original image using wavelet transform and according to the specific texture image, select the appropriate wavelet sub-image (approximate image or its detailed sub-image) for texture analysis. The texture feature parameter (contrast) of the decomposed image is utilized to determine the optimal step parameter. When the step parameter is optimal, the texture feature parameter reaches the extreme value of the period which is beneficial to texture analysis. As the amount of data in the decomposed image is less than that in the original image, both the computation complexity and the time consumed in finding the optimal step parameter are reduced. Furthermore, experimental results show that the optimal step parameter of the wavelet decomposed image is accurate.