The run speed of a system of natural language processing is an extremely important bearing on developing the system itself. One of the cores of the system is a grammar parser. This paper gives the general sequencial algorithms of simple backtraking, thorough backtracking and pruning, which are used in a grammar parser, and some thinking about implementing these algorithms. Our experiments and analysis show that these algorithms are very efficient.