The increasing use of 3D models in many areas lead us to think about the impact of the different distortions that can affect the 3D object during the rendering process. These deformations are usually evaluated using geometric metrics, which have not a good correlation with human judgment, while the visual perceptual quality of 3D models is necessary. In this context, a new full-reference metric denoted LWRMS is defined in this work. It can predict the distortion score between the original object and its damaged version without taking into account the constraint of connectivity. The proposed metric is defined in order to get a good correlation with the human visual perception coming from a subjective measurement. The numerical experiments are carried out on a known database LIRIS/EPFL General-Purpose database. The quantitative results show a good performance of the proposed metric in comparison with methods from the literature.