Composite services usually run on the open, ever changing and uncertain Internet. Dynamic and complex execution environments make the composite service execution uncertain and unreliable, undermining the practical usability of service composition techniques. To solve this problem, an EQ rule based composite service runtime adaptation approach is proposed from the perspective of environment analysis and modeling. In this approach, the composite service execution log data and the candidate service execution environment data are first collected. Then, the EQ rules which express the knowledge (e.g. “how the performance of one candidate service will be in specific execution environment state”) are mined. At last, the discovered EQ rules are applied to response all kinds of real time generated environment change events and to drive composite service runtime adaptation, and as such to ensure reliable execution for composite services. Experimental results show that the proposed approach can improve the quality of composite services and increase their execution stability effectively in dynamic environments.