In SOA, one challenging problem is the QoS evaluation for the newly composed service. Most existing work on service composition focuses mainly on the static QoS behavior of individual components. Many other factors such as execution profile (e.g. service task and network dynamics) and "best results" with slow selection time vs. "good enough" results with fast selection time are often ignored. To address the practical aspect of service selection and to perform more appropriate tradeoff, we propose a service QoS model and evaluate two selection algorithms in this paper. Experimental results show that our model and mechanism can provide good tradeoff between service selection time and QoS metrics.