Event anomaly detection in dynamic graphs, such as the identification of abnormal transaction behavior on e-commerce platforms, focuses on detecting anomalous activities as the graphs evolve over time. While many approaches have demonstrated significant advantages in managing complex graph structures, relatively few have addressed the core challenges in dynamic graph anomaly detection, namely the heavy reliance on labeled data and the insufficient mining and integration of spatial and temporal information. In this paper, we present DPEAD, a method for detecting event anomalies in dynamic graphs, which uses a pre-training and prompt fine-tuning framework. First, a self-supervised pre-training approach is employed to alleviate the scarcity of labeled data. Second, a dual-prompt mechanism for anomaly detection is developed to improve the integration and utilization of spatiotemporal information, effectively bridging the gap between pre-training and downstream tasks. Specifically, DPEAD utilizes both snapshot-view and event-view approaches to model dynamic graphs at different stages, enabling the effective extraction of dynamic information. Our experimental results, based on four real-world datasets, demonstrate that DPEAD achieves superior performance compared to existing state-of-the-art approaches.