Scale-free connectivity, small-world pattern, hierarchical modularity and disassortative mixing are prominent features shared by most biological networks. Up to now, various network growth models invoking gene duplication and divergence have been proposed to understand the evolutionary mechanisms shaping the scale-free connectivity, small-world pattern and disassortative mixing . In this paper, we present an evolutionary model by introducing a rule of small preference duplication of a node and sequently divergence plus non-uniform heterodimerization meaning that the probability that an heterodimerization link is added between duplicated nodes is proportional to the number of common neighbors shared by these nodes based on biological background, showing that our model can almost display series of topological characteristics of real protein interaction networks, such as scale-free connectivity, small-world pattern, disassortativity of degree-degree correlation and hierarchical modularity. Our model may yield relevant insights into the evolutionary mechanism of protein interaction networks behind it.