Functions and call relations are extracted into nodes and edges, respectively, by which a novel topological model is proposed. The directional edges and the corresponding weight values express the call relations and tightness degree. By introducing two concepts of function fault-tolerant capability and software fault intensity and by designing a allocation rule on fault-tolerant capability, a cascading fault model is built to explore fault propagation behaviors. Simulations on practical software networks show that a weak fault intensity, a small number of initial faults, and a strong fault-tolerant capability can slow down the spreading speed. The functions with more call relations and more closer tightness contribute more to the stability of the whole system.