Abstract: The recently proposed Zest method could solve the problem that test inputs generated by ordinary fuzzy methods often fail to pass the checks inside structured input programs. However, Zest cannot make full use of existing input cases to improve code coverage like popular coverage-guided fuzzy systems (such as AFL) do. This paper proposes Zestlet, a fuzzy test method for structured input program integrating fragment information. It uses a fragment generator to obtain fragment information from the example inputs, and then uses a parametric generator that integrates fragment information to generate inputs during fuzzy test. This paper tests Apache Ant, Maven and Tomcat and confirms that Zestlet improves both total coverage and valid coverage compared with Zest, and also reduces the time required to achieve the same coverage by 69% to 99%.