It’s very difficult to find an appropriate parking space in urban area, when drivers are near to their destinations. The literature studies showed that 30% of the traffic congestion, unnecessary fuel consumption and exhaust emissions are caused by searching for the parking spaces. With the ever-changing nature of technology, smart parking systems composed of smart devices and sensor technologies are readily available and provide various information such as locations, real-time available counts, and parking costs. However, the drivers can’t know whether there is an available parking space at the arrival time. In this paper, we propose a parking occupancy prediction approach based on spatial and temporal analysis. In this approach, we extract related features and build the parking occupancy prediction model by Naïve Bayes classifier and decision tree. The prediction model can be used to predict the level of parking occupancy rate for each street block in the next hour. To evaluate the performance of proposed approach, we carried out the experiment by the on-street parking data collected by the SFPark system in San Francisco, USA. The results show that our proposed smart parking guidance system can significantly improve the prediction accuracy for the level of parking occupancy rate.