We focus on relation extraction, which classifies a relation given a sequence containing two pre-defined entities. Recently, as contextual pre-trained language models have shown significant performance improvements in various natural language processing tasks, they have been adopted in the relations extraction task. in this paper, we explore several methods of classifying the relation: 1) Standard, 2)Mention Pooling, and 3) Entity Start. Evaluation results on a large Korean relation extraction corpus show that Entity Strat method is highly effective to extract relation between two entities.