Copyright protection becomes a more and more important problem with the development of the creation and distribution of 3D model. The paper proposes a novel and robust watermarking scheme for 3D model based on the feature recognition method. Firstly, we construct the Cartesian coordinates in 3D model and extract six 2D images from six different directions according to the habit of human recognition. Secondly, We divide the images into 8?8 blocks and transform them in DCT domain, then embed the watermark into the middle-frequency coefficients according to the JND model of the human visual system. Finally, we retrieve the 3D model which is embedded the watermark into. The watermarking detection process is averse to the watermarking embedding process. Many experimental results show that this watermarking scheme is imperceptible, robust and able to withstand common attacks on 3D models such as mesh simplification, adding noise, model cropping.