The invention discloses a fault-tolerant method for constructing a data center network based on a cross-cube network, which belongs to the cross-technical field of computer and mathematics. The present invention comprises the following steps: according to the definition rule of CQn, generates the association matrix code of CQn, obtains the adjacency point of each vertex of CQn, saves the adjacency relation between the vertices in the association matrix, generates error set afterwards, eliminates in the association matrix Error set, and finally perform depth-first traversal on the remaining sub-matrix to find the required path, that is, to judge whether there is a cycle of a specified length in the matrix and record a qualified cycle. The remaining sub-matrix is the part of CQn after removing the error subset. The present invention adopts the cross cube network CQn to construct an efficient, fault-tolerant and scalable data center network, and its fault-tolerant capability is n-2, that is, when n-2 errors occur in the network, the rest of the system can still be guaranteed to operate normally , which improves the fault tolerance capability of the system, and also improves the topological performance such as scalability and reliability of the data center network.