Abstract: The current mainstream intelligent Fuzzing often constructs new test samples through precise analysis of the programs internal structure, which is heavily dependent on the performance of the computer and often overlooks the guiding significance of the discovered program information of exceptions for construction of new testing samples. To overcome these shortcomings, this paper presents a method based on intelligent Fuzzing exception distribution steering, which establishes a data-constructing model named TGM (Testcase Generation Model) for binary program testing. Firstly the relevant information of testing samples is collected according to the computing capability. Then random initial testing samples are selected for testing. Finally, the testing results are used to initialize parameters of the model, which guides the priority selection of more effective input attributes to construct new samples for the next round of testing. This procedure is repeated in iterative testing to constantly update model parameters for guiding the next testing. Experimental data shows that this method can assist Fuzzing to prioritize more effective samples for testing. Design prototyping tool CombFuzz has good performance in the exception detection capability and code coverage capability, meanwhile, when the tests are carried out on large programs, compared with MiniFuzz of Microsoft,s SDL lab, this method increases the average of exception detection rate by nearly 18 times in a limited period of time, and has found 7 undisclosed exploitable vulnerabilities in WPS 2013 and other softwares that MiniFuzz did not find.