The phase measuring deflectometry (PMD) has attracted extensive attention to researchers in recent years as it has the advantage of being fast, non-coherent and high sensitivity. The high sensitivity of PMD allows measuring gradi-ent changes in the range of micro-scale and local height changes in the range of nanometers, which enables PMD as an effective tool for high precision inspection of defects or local height variation. The accuracy of the PMD is related to the phase reliability of the captured fringe pattern. Errors in the phase map influence the accuracy of the whole measurement. When testing low reflectivity specular surfaces like cell phone shell, contrast of the distorted fringe patterns is low, and there are always relatively big errors in the phase map. The phase error characteristics in PMD system when testing low reflectivity surfaces are analyzed. The results illustrate that the random phase error increas-es rapidly while the nonlinear error drops slowly with the decreasing of the tested surface reflectivity. In order to at-tain high precision measurement of low reflectivity specular surface, a robust error reduction method based on wavelet de-noising is presented to reduce the phase error. The optimal wavelet parameters for denoising the aimed noise level are carried out by simulation, which are 5 decomposition level, ‘soft’ thresholding and ‘rigrsure’ thresh-olding rule. The error reduction method is compared with the least-square TPU method and low-pass Gaussian filter method. As the result, when compared to the least-square TPU method, the method based on wavelet de-noising needs much less shooting time and has a more outstanding error reduction effect. In comparison with the low-pass Gaussian filter method, the wavelet de-noising method performs better in the preservation of phase details. The ex-periment of measuring a typical mobile shell shows clearly the superiority of the method based on the wavelet de-noising. In some situations, if the curvature maps are required for the inspection of defects, especially when the tested surfaces have low reflectivity, the method based on wavelet de-noising would be quite suitable for error reduc-tion. The method based on wavelet de-noising is also suitable to detect small defects and for the measurement of the high reflective surface to reach higher precision. In the experiment of measuring a plane mirror, the RMS phase error with the method based on wavelet de-noising is 5 times smaller than that with only 6-step phase shifting method.