Aiming at the candidate set construction problem in entity linking systems, we propose a multistrategy candidate set construction method. We use a variety of strategies to extract the complete mentions in the context, reduce the number of candidates and improve the recall of the correct entity in order to construct a high quality candidate set. We conduct experiments on the TAC2014 English entity linking corpus and then analyze the results. The optimal construction strategy is chosen. Experimental results prove that the proposed method can improve the recall and precision of candidate sets. We further validate that the quality of the candidate sets has a significant effect on the performance of the whole entity linking system. Compared with the baseline algorithm, the candidate set extracted by the optimal candidate set construction strategy can improve the performance of the whole entity linking system by 3.7%.