Zero-shot relation extraction is proposed to address the issue that the model performance drops dramatically when identifying novel relations that cannot be observed at the training stage. Existing works focus more on the pattern of matching between instances and relation descriptions, however, they ignore the capabilities of pre-trained model in zero-shot scenarios. In this paper, we propose a novel prompt-based joint contrastive learning framework (PCL) to enhance zero-shot relation extraction. Specifically, we construct a prompt template to leverage prior knowledge in PLMs for better representation of the relation. In order to learn more effective features, we design two contrastive learning modules to jointly optimize the obtained representations. Experimental results on two publicly available datasets show that PCL outperforms the previous methods.