With emerging of multiple services brought by smart power grid, the rapid evolution of 10G-EPON (10 Gigabit Ethernet Passive Optical Network) provides great access for multiple services with high capacity. However, newly emerging multiple services have diversified characteristics and demands, which presents great problems for current 10G-EPON to match these services with high quality and matching degree. To satisfy requirements of diversified types of services in power electirc system, services charactrisetcs and demands of smart power grid are discussed, and a Bayesian Classifier based Service Awareness (BC-SA) mechanism of 10G-EPON is proposed for smart power grid. According to the “master-slave” architecture between OLT (Optical Line Terminal) and ONUs (Optical network Units) in 10G-EPON, a similar “master-slave” structure is adopted and designed in the proposed BS-SA mechanism. By using Bayesian classifier, the BC-SA mechanism is able to be aware of the type of service. On the basis of Bayesian classification, resources allocation and transport policy are both determined by the proposed mechanism throught cooperation between OLT and ONUs. In order to verify the reasonability of this BC-SA, system simulation is conducted in domains of delay and packet-loss-rate. Simulation results show that the BC-SA is able to achieve greatly better performance of both delay and packet-loss-rate, and improve the supporting ability for data services of smart power grid with better matching degree.