A few of works have recently focused on relieving service users of the role played in assigning numeric values to QoS criteria as required in traditional service selection scheme. Towards such goals, skyline services for example have been studied recently by a lot of researchers. However, the size of skyline services sometimes is not easily controlled due to intrinsic attributes of services. Besides, some QoS metrics, such as response time, reliability, and etc., usually suffer from network as well as Internet factors (e.g. network unavailability and Internet disconnection), which contributes to the fact that these QoS metrics may fluctuate during run time. Considering this kind of dynamics of QoS metrics, we in this paper propose to obtain probabilistic top-k dominating composite services with uncertain QoS. More important, the number of obtained services can be easily specified by service users, thus avoiding the case that the size of skyline services is sometimes out of control. The experimental results have investigated the feasibility and effectiveness of our approaches.