The prediction of protein-protein complex structures is crucial for fundamental understanding of celluar processes and drug design. Despite significant progresses in the field, the accuracy of ab initio docking without using any experimen-tal restraints remains relatively low. With the rapid advancement of structural biology, more and more information about binding can be derived from experimental data such as NMR experiments or chemical cross-linking. In addition, informa-tion about the residue contacts between proteins may also be derived from their sequences by using evolutionary analysis or deep learning. Here, we propose an efficient approach to incorporate interface residue restraints into protein-protein docking, which is named as HDOCKsite. Extensive evaluations on the protein-protein docking benchmark 4.0 showed that HDOCKsite significantly improved the docking performance and obtained a much higher success rate in binding mode predictions than original ab initio docking.