Network protocols define the data communication paradigm for entities during interactions and serve as the cornerstone of today’s Internet architecture. However, implementations often require compromises in performance, latency, portability, etc. that can introduce security risks. In order to uncover vulnerabilities of the server, network fuzz testing is in general use, which entails continuously generating diverse data messages and simulating communications with the entities under test. Current network fuzz testing methods often demand extensive prior knowledge when addressing unknown protocols. In this paper, we proposed a generalized fuzzing framework, UNPfuzz. Based on deep learning methods and PinTools, we can effectively simplify the process, reduce the need of the professional knowledge and be independent from the prior information.