The methods and progress of knowledge-based protein three-dimensional structure prediction are mainly introduced. To date, the methods of knowledge-based structure prediction can be grouped into two categories. One is homologous protein modeling, which is relative mature and can give reliable results, but it is restricted to modeling of the target sequence that shares a high sequence identity with thehomologous proteins. The other method, inverse protein folding, can be empolyed to predict the structure of protein with limited sequence homology, which mainly includes 3D Profile and potential-based functions, and gives the topology of the target protein.