Several approaches have been proposed to deal with the web service matchmaking problem. Unfortunately, most of these solutions are purely syntac- tic measures based on the input/output interface specifications of web services and consequently lake accuracy. This is a serious drawback in a fast growing In- ternet that is facing the challenge to deal with an increasing number of published services. The proposed solutions to cope with this limitation consider the pro- cess part of a service description as a graph in the similarity measure. This kind of solutions has better accuracy but suffer from high computational complexity because they rely on time consuming graph matching tools. To avoid this heavy time computing overhead, we propose in this paper a solution that decomposes the process graph into smaller subgraphs and construct similarity of web services based on the similarity of their subgraphs. Simulation results show that our solu- tion is both accurate and fast to compute.