Predicting the cost of environmental governance is an essential process in environmental protection. However, the existing cost prediction methods face several challenges, including the necessity of considering the causality of environmental governance, the importance of distinguishing environmental indicators, and the difficulty of collecting environmental data. In order to address these challenges, a novel rule-based system, called the extended belief rule-based (EBRB) system, is first introduced to establish the basic framework of cost prediction. Then, a combination of structure learning and parameter learning, or joint learning, is developed to improve the performance of the EBRB system. Finally, a new cost prediction method based on the improved EBRB system is proposed for environmental governance. To verify the effectiveness of the new cost prediction method, an experimental study is carried out to compare the predicted cost of environmental governance in 29 provinces of China. The comparative analyses demonstrate that the new cost prediction method can not only provide a desired level of accuracy, but also exhibit excellent robustness that makes it better than some existing cost prediction methods.