Based on Application Behavior Declaration (ABD), we proposed a new method for credibility calculation. It is based on the idea of "words and deeds". In the software development phase, tracking module is implanted to extract the action path of the behavior. Then, we compare the behavior declaration of the software with the action path to achieve the purpose of evaluating whether the behavior is credible. Explicit and implicit indicators are proposed for judgment, as well as an implicit indicators model based on K-means clustering. This model is applied to the credible calculation of single behavior and the credible discrimination of similar behaviors. Experiments prove that this solution is feasible for providing new ideas for credibility testing.