Programs using service-oriented architecture (SOA) often feature ultra-late binding among components. These components have well-defined interfaces and are known as web services. Messages between every pair of web services dually conform to the output interface of a sender and the input interface of a receiver. Unit testing of web services should not only test the logic of web services, but also assure the correctness of the web services during input, manipulation, and output of messages. There is, however, little software testing research in this area. In this paper, we study the unit testing problem to assure components written in orchestration languages, WS-BPEL in particular. We report an empirical study of the effectiveness of the Frankl-Weyuker data flow testing criteria (particularly the all-uses criterion) on WS-BPEL subject programs. Our study shows that conventional data flow testing criteria can be much less effective in revealing faults in interface artifacts (WSDL documents) and message manipulations (XPath queries) than revealing faults in BPEL artifacts.