The complex networked computer systems of today are often vulnerable to a large number of failures, accidental as well as intentional. To be able to assess to what degree one can rely on such a system, new methods for quantitative evaluation is needed. This paper presents a stochastic model for integrated security and dependability evaluation, which models malicious attacker behavior as transitions between system states. To predict the probabilities of attack actions a game theoretic approach is applied. We demonstrate the method by computing security and dependability measures in two different case studies.