With the advanced Internet technology, business applications across multiple enterprises based on Composition web services (CWS) paradigm are widely used. Since business processes among enterprises become complex, loosely coupled, long running and unpredictable, tasks collaborate in a peer-to-peer fashion without central control, in which task dependency is inevitable. And when a system crash occurs, some tasks of the transaction flow may be committed while others unscheduled, in this situation, it is important to accurately specify the dependency between tasks in Transactional composition service (TCS), contemporary technologies usually statically specify dependency point and avoid implicit interaction in parallel aggregation. In this paper, we propose a task dependency analysis method based on transaction execution logs. We present the formal description of atomicWeb service (WS) and discuss properties of the atomic WS such as pivot, compensability, retriability and vitalness, then analyze execution logs of transactions, specify data flow dependency and behavior dependency between tasks in TCS, behavior dependency and properties of aggregation patterns. The TCS based application of Trip reservation process (TRP) shows that it is feasible to ensure consistent execution of reliable TCS.