Traffic prediction is a fundamental research problem in traffic planning and management, with wide-ranging applications. The task is challenging due to: (1) Most methods in traffic prediction emphasize short-term forecasting tasks and encounter difficulties in accurately modeling both short-term and long-term dependencies simultaneously. (2) Existing methods do not delve deeply enough into capturing spatial dependencies and lack sufficient exploration of spatial information in traffic data, resulting in suboptimal effectiveness. To address these issues, this paper proposes the Multi-views Enhanced Spatio-Temporal Adaptive Transformer(MESTformer) for Urban Traffic Prediction. Specifically, the network designs spatial attention and temporal attention to explicitly model the spatio-temporal dependencies. And two spatial enhanced views to delve into the spatial information of traffic data from local geographic and global sematic perspectives. By combining spatial enhanced views with spatial attention, the model can more fully learn spatial features. In addition, we manually insert uniform attention between the two ST blocks to ensure coverage of a broader range of information. We conduct extensive experiments on four prominent traffic benchmark datasets to evaluate the effectiveness and efficiency of MESTformer. The experimental results indicate that our proposed MESTformer achieves state-of-the-art performance compared to competitive baselines.