The self-similarity of complex has extensive practical background in real world. It is similar to the phenomena of social relationships: “things of one kind come together, birds of a feather flock together”. Hence, we proposed the self-similarity network evolving model based on attributes similarity between the nodes. In network each node has the attribute value, by this computed similarity between the nodes. If two nodes attribute similarity falls in certain sector, then established the connection between nodes. The simulations make clear that the degree distribution of the self-similarity network similar to the small-world networks. The clustering and the average path of the self-similarity network are smaller than BA model and are bigger than small world. Similarity network model is a new characteristic of complex network except the BA mode and the small world mode. On the other hand, the self-similarity network has good robustness to random fault and deliberately attack.