The creation of value-added composite web service raises an opportunity to the formation of online B2B collaborations. However, web services are usually overlapping in functionality. How to make a choice based on non-functional factors becomes a problem that needs to be solved. In this paper, it argues that the selection of component services should be considered in a global manner and the user's QoS preferences is more appropriate to be expressed in a fuzzy way. It advocates using fuzzy set to express the user's QoS preference on a specific QoS criteria and using fuzzy expression to express the user's trade-off among QoS criteria's. The service selection problem is formalized as a fuzzy constraint satisfaction problem and deep-first branch-and-bound method is chosen to search for a solution with some adjustments to web service composition. It also develops an uncritical fuzzy consistency checking algorithm as an aid to faster search. Experiments are conducted to prove the effectiveness of the approach.