Trust measurement, which is the basis of "measurement, storage, and reporting" of trusted computing, is still lack of mathematical theory and has few applications in a real-time environment thus far. The difficulty lies in three points. One is how to establish a general mathematical model that can cover different mainstream definitions of "trust"; the second is how to build a runtime trust measurement theory based on the established mathematical model; and the third is how to map the above the model and theory to real information systems, and therefore form a practical real-time measurement method. To address the above issues, a runtime software trust measurement approach is proposed. Initially, a noninterference model is leveraged to explain different mainstream definitions of trust, indicating that noninterference model can be an appropriate option of general mathematical model for trusted computing. Next, a noninterference model-based real-time trust measurement theory is presented. In the proposed trust measurement theory, a system call is processed as an atomic action, and the sequence of system calls is constructed as the real behaivior of a process. Note that every system call belongs to a security domain, and different security domains are of noninterference with each other. Therefore, after obtaining a real behavior α, the theoretically expected behavior β can be calculated based on the noninterference relations between security domains to which system calls in α belong. Once obtaining α and β, the trust of a process can be measured by determining whether two behaivors α and β deviates. Finally, a trust measurement algorithm is given. The algorithm can determine whether a process trust or not, i.e., whether the real behavior α and the theoretically expected behavior β deviates, within the time complexity of O(1). The proposed theory is also applied into real information system, and experimental results show that the proposed approach is effective and efficient.