Abstract: With the widespread use of agile software development methods, such as agile and scrum, software is iteratively updated more frequently. To ensure the quality of the software, regression testing is conducted before new versions are released. Moreover, to improve the efficiency of regression testing, testing efforts should be concentrated on the modified and impacted parts of a program. However, the costs of manually constructing new test cases for the modified and impacted parts are relatively expensive. Fuzz testing is an effective method for generating test data automatically, but it is usually devoted to achieving higher code coverage, which makes fuzz testing unsuitable for direct regression testing scenarios. For this reason, we propose a fuzz testing method based on the guidance of historical version information. First, the differences between the program being tested and the last version are analyzed, and the results of the analysis are used to locate change points. Second, change impact analysis is performed to find the corresponding impacted basic blocks. Finally, the fitness values of test cases are calculated according to the execution traces, and new test cases are generated iteratively by the genetic algorithm. Based on the proposed method, we implement a prototype tool DeltaFuzz and conduct experiments on six open-source projects. Compared with the fuzzing tool AFLGo, AFLFast and AFL, DeltaFuzz can reach the target faster, and the time taken by DeltaFuzz was reduced by 20.59%, 30.05% and 32.61%, respectively.