Service discovery is a critical stage of the development for Internet-scale software produced through service composition. Presently, development efficiency of composite service is confined by a low degree of automation and accuracy of service discovery. This paper propose the AutoDisc (automatic service discovery framework based on business process similarity) schema to improve development efficiency through two aspects. One is to improve the efficiency of discovery by automatic recommendation. The other is to improve the accuracy of service discovery by combining the structural and behavioral factors of services. Through the proposed approach, this paper automatically models the requirements of service discovery and recommends the most appropriate composite services to developers. Finally, the paper illustrates the effectiveness of AutoDisc with a set of experimental evaluations which show that AutoDisc can increase the development efficiency by 75.5% in the evaluating context.