A novel blind watermarking method for the 3D mesh model is proposed. This method is based on the geometry image which is created by the parameterization and sampling of the model. The geometry image except the boundary is partitioned into several sub-images of equal size. Every sub-image is normalized and decomposed with two-dimensional Haar wavelet. The watermark is embedded into the low components of the wavelet coefficients with the LSB method. Other conventional 2D image watermarking methods could also be applied. The watermarked model is reconstructed from the embedded geometry image. For the extraction of the watermark, the original model is not required. Experimental results show the proposed algorithm is robust against the attacks such as translation, scaling, rotation, local deformation and certain amplitude of noise disturbance.