Abstract: The development of information technology makes the government and military authorities put forward higher requirements on the security of information system. The uniqueness of classified information system security makes risk assessment different from the common information security system. In this paper, classified information systems are research object. We first describe the characteristics of information system security, the uniqueness of the existing methods of analysis and evaluation of information security risk assessment. Then an evaluation model based on AHP is introduced into security risk assessment of information system security. A new way of classified information system risk assessment is presented. At last, through the analysis of an example, we analyze that the proposed model in processing of information system security evaluation process to get the discrete data distribution is not required, compared with the Delphi method and BP neural network, this model is highly of practicability and expansibility, it does suitable for classified information system risk assessment actually.