Abstract: With the increasing integration of information and communication technology, the cyber-physical power system (CPPS) is facing not only uncertainties from the physical world but also threat from cyberspace. It is urgently needed for a method to assess the defensive capability of CPPS against cyber attacks. An assessment method is proposed for intrusion tolerance and optimal defense resource allocation of CPPS based on mean failure time and reliability. Firstly, a semi-Markov chain attack model is established for the advanced persistent threat (APT) to analyze the penetration process of attacks from cyber level to CPPS. Besides, a stochastic game model is adopted to dynamically describe the interaction process between attackers and defenders in CPPS, subsequently predicting the optimal strategy of attackers under Nash equilibrium, and determining the optimal defense strategy against malicious attacks. Finally, the effectiveness of the intrusion tolerance assessment method is verified through a CPPS security testbed, which shows that the intrusion tolerance capability has a significant role in the secure operation of the CPPS.