Projection data obtained through optical techniques for tomographic measurement are often with noise. The reconstruction accuracy is deteriorated by the Gauss- or Poisson-distributed noise. Corresponding to the major steps of tomographic technique, a two-stage denoise technique has been developed. The deflectometric data are denoised by wavelet-based procedure after projection data reduction. Deflection-angle averaging revision algorithm is applied to suppress the noise in iteration. The Moire tomographic reconstruction technique with eight-neighbor derivative algorithm and denoise method is tested using numerically generated data from a known temperature distribution and noise source. The results show that the denoise processing is able to improve the reconstruction accuracy significantly, which reduces the reconstruction errors of 45%. The temperature field of butane flame is then measured by the Moire tomographic technique and the experimental reconstruction is compared with the thermocouple temperature measurement.