The invention discloses a multiprocessor network fault node diagnosis method based on a folding hypercube. The multiprocessor network fault node diagnosis method comprises the following steps: S1, determining a symptom set omega generated by testing between folding hypercube network FQn nodes; s2, generating a 0-comparison sub-network omega0 (FQn) according to the symptom set omega of the FQn; s3, the maximum connected branch CCLarge of the 0-comparison sub-network omega0 (FQn) is obtained through depth-first search; s4, diagnosing all the nodes in the CCLarge as no fault, and adding the nodes in the CCLarge into a fault-free node set SFault-free; and S5, diagnosing the remaining nodes in the folded hypercube network FQn except the fault-free node set SFault-free, and obtaining the final fault-free node set SFault-free and the fault node set SFault. According to the method, the network topology structure property of the folding hypercube network FQn is utilized, and the fault diagnosis capability of the FQn is greatly enhanced under the condition that a small number of nodes are misdiagnosed.