The precise characterization of surface morphology is essential for predicting the joints quality. Compared to the conventional two-dimensional profiling, the full-surface 3D profiling captures more comprehensive spatial data which is high related to the joints mechanical performances. Therefore, a method for full-surface 3D profiling and multi-dimensional feature extraction is proposed for the intelligent prediction of weld quality in laser welding. A high-precision 3D model of the entire weld surface without geometric blind spots is efficiently reconstructed utilizing a multi-view vision acquisition system and point cloud registration. Multi-dimensional geometric features are quantitatively extracted from the full-surface 3D profile of weld. It is demonstrated that the complex weld features can be captured by the full-surface profiling method effectively. Furthermore, the intrinsic relationship between the multi-dimensional weld features and the mechanical performance is accurately established by the intelligent algorithm. The results indicate that the identified key weld features facilitate the construction of a highly robust intelligent prediction model. The extracted weld features from multi-dimensions reveal spatial variation patterns of the weld which are extremely difficult to be analyzed using traditional two-dimensional profiles. Therefore, the proposed full-surface 3D profiling and multi-dimensional feature extraction method is of great importance for quantitatively characterizing the weld morphology and realizing the intelligent prediction of laser weld quality.