The space information networks are easy to be disturbed and destroyed, which is due to their hostile operational environments and results in a decrease in network quality of service. Given their characteristics, a distributed topology failure diagnosis scheme composed of failure detection node election and failure diagnosis is proposed. First, a proper number of nodes are elected to detect faults in a coordinated and distributed way by using minimum dominating set algorithm. Then elected diagnosis nodes detect faults around based on the theory of dependency-graph. The technology of minimum Huffman tree is used to preliminarily locate failures and on that base the further location is fulfilled through sending detection messages on purpose. Simulation results from NS2 demonstrate that high accurate diagnosis rate can be reached at the expense of low cost by this failure diagnosis algorithm.