A novel attack method on robust 3D mesh object watermarking applications was proposed. Here we propose an adaptive noise of watermarking attack, which can make the noise intensity and the noise direction adjust in different parts of 3D mesh object. The proposed attacks consist of two main stages. First, adjust the 3D mesh object in the direct neighborhood, which can ensure to remove the watermark which watermark intensity is out of the direct neighborhood. Second, different regions of object use different smoothing method to filter. Third, add adaptive noise to the 3D mesh object of data points and in the HVS human visual system constraints adaptive noise has always been maintained a greater intensity noise. The results show that the watermark attack algorithm is better than other similar algorithms in the average performance.