Knowledge-enhanced methods that take advantage of auxiliary knowledge graphs recently emerged in relation extraction, and they surpass traditional text-based relation extraction methods. However, there are no unified public benchmarks that currently involve evidence sentences and knowledge graphs for knowledge-enhanced relation extraction. To combat these issues, we propose KGRED, a knowledge graph enhanced relation extraction dataset with features as follows: (1) the benchmarks are based on widely-used distantly supervised relation extraction datasets; (2) we refine these existing datasets to improve the data quality, and we also construct auxiliary knowledge graphs for these existing datasets through entity linking to support knowledge-enhanced relation extraction tasks; (3) with the new benchmarks we curated, we build baselines in two popular relation extraction settings including sentence-level and bag-level relation extraction, and we also make comparisons among the latest knowledge-enhanced relation extraction methods. KGRED provides high-quality relation extraction datasets with auxiliary knowledge graphs for evaluating the performance of knowledge-enhanced relation extraction methods. Meanwhile, our experiments on KGRED reveal the influence of knowledge graph information on relation extraction tasks.