A new benchmark, CapMem, is introduced for evaluating episodic memory in egocentric video. The benchmark consists of 75 videos with 1,000 multiple-choice questions, and results show that caption-based memory outperforms direct video QA on long videos.