The invention belongs to the technical field of network security, and particularly relates to a cross-architecture binary function similarity detection method and system based on a neural network, andthe method comprises the steps: traversing a binary file function list for different types of binary files, constructing and optimizing a function control flow graph, translating the program basic block byte code to obtain an intermediate representation, generating a semantic embedding vector of the basic block code, extracting function control flow graph nodes by using a breadth-first graph traversal algorithm, obtaining function embedding vectors according to semantic embedding vectors and control flow information of the nodes, and calculating the cosine distance between the function embedding vectors to measure the function similarity. The method is more beneficial to code intermediate representation, eliminates the difference between different instruction architectures, reduces the cross-architecture code similarity detection difficulty, and reduces the expansion workload and difficulty, based on a function embedding process of a PVDM model and a graph neural network, introductionof human prejudice is avoided, the improved graph neural network is faster in convergence speed and higher in overall efficiency and accuracy of the system.