Testing is the traditional validation method in the software industry. To ensure the delivery of high quality and
robust service-oriented applications, testing of web services composition has received much attention. These services have become more and more complex, where they have to cope with strict requirements of business processes and their
dynamic evolution, and interactions among different companies. In this context, the analysis and testing of such services
demand a large amount of effort. To reduce the effort required for web-services testing, in this paper, we propose a
specification-based approach to automatically generate test cases for web services composition that is modeled at different levels of abstraction. This approach specifies a service structure as multi-level models. To generate the test cases, it checks if the first level of the model has a parallel execution or a decision table to be solved by an algorithm that solves Chinese postman problem. Then, it identifies paths for last level of the model and relates the results of all levels with each other. To evaluate our approach, we applied it to four cases study using our developed tool. Compared to exiting approaches, our approach reduces testing cost and execution time, and increases testing reliability.