This paper shows two power analysis attacks against a software implementation of a first-order DPA resistant S-box algorithm that is based on the discrete Fourier Transform (DFT). The DPA resistant S-box algorithm based on DFT was proposed by Prouff et al. in 2006 and improved by Coron et al. in 2008, respectively. In our attacks against the improved one, we pre-process the power traces by separating them into two subgroups, so that each has a biased mask. For the separated power traces, two post analysis methods are proposed to identify the key. One is based on DPA attack against one subgroup, and the other utilizes the difference of means for two subgroups and a pattern matching. Finally, we compare these two attack methods and propose an algorithm-level countermeasure to enhance the security of S-box calculation based on the DFT.