Abstract: As an emerging subfield of quantum machine learning, quantum deep reinforcement learning (QDRL) utilizes quantum neural networks (QNNs) to construct a quantum agent and trains QNNs through multiple interactions with an environment to maximize the expected cumulative return. However, existing QDRL methods require the quantum agent to interact with a classical environment many times, requiring a huge number of executions of the QNN circuit. To address this problem, this work proposes a QDRL model, a quantum episodic memory deep Q-network, which utilizes episodic memory to accelerate the training process. Specifically, the proposed model stores experiences with high rewards in history into the episodic memory, which then helps the quantum agent to obtain the desired action with significantly fewer iterations when the environment state is similar to one of those stored in the episodic memory. Numerical simulations on five typical Atari games show that the proposed method can significantly reduce the number of training iterations and can achieve a higher score compared to other conventional QDRL methods.