The trusted computer can only ensure the static security at present. When the system begins to run, the trustworthiness of terminal computer mainly depends on the trustworthiness of software behavior running on it. Therefore, the dynamic trustworthiness of software during running has become a critical problem to solve the trustworthiness of computer system. According to the dynamic trustworthiness of software during running, checkpoint's scene-level attributes based on interval data are introduced and a dynamic trustworthiness evaluation model of software based on checkpoint's classification attributes is presented based on the traditional software behavior model. The model evaluates the trustworthiness of software behavior based on the trace of expected behavior. It uses checkpoint attributes' classification strategy which simplifies the trustworthiness evaluation of checkpoints based on threshold, and combines checkpoint attributes' subjective classification weighting with scene-level attributes' objective weighting. The model evaluates the trustworthiness of scene-level attributes based on interval data by constructing trusted model of scene-level attributes. The theoretical analysis shows that the classification attributes decision making reduces the number of comparisons with the threshold, and the experimental results verify the effectiveness of trustworthiness evaluation of scene-level attributes based on interval data and the high attack detection capability of the dynamic trustworthiness evaluation model.