Taking entity as the basic unit to organize the tuples in query processing is an effective way of managing low-quality data. As many descriptions of attribute value in an entity, join operator must support similarity join over multiple values. Entity similarity join is more effective than traditional similarity join in data cleaning, information integration, fuzzy keyword search, fraud detection, and text aggregation. In this paper, an entity similarity join algorithm, ES-JOIN, is designed by adopting the structure of the double layer prefix index. The presented method is suitable for solving set similarity join problem based on fuzzy elements matching, and thus is a better choice than the traditional set similarity join which only considers exact element match. In order to accelerate the join process, a new filtering measures is proposed to optimize the algorithm, and an optimization algorithm, OPT_ES-join, is also obtained. Experiments demonstrate that the ES-JOIN algorithm has good efficiency and scalability, and the filter measures is very effective.