While a lot of Web Services are available nowadays, Quality-of-Service (QoS) becomes a crucial concern to distinguish alternatives from each other. QoS-based service composition aims to determining optimal combinations among exponential candidates with consideration of overall non-functional performance. In this paper, a multiple objectives optimization approach is proposed for the decision-making process. First, we design a pattern-wise replacement method to derive composite QoS objectives of a given XML-encoded workflow process. According to the process descriptions, the composite QoS objectives will be available by synthesizing non-functional metrics of primitive patterns repeatedly. Second, we employ the ε-Pareto dominance relations to discriminate the overall QoS performance of combinations. Thus user can freely set any desirable ε values to define the precision of service composition’s QoS performance. Finally, we present a genetic algorithm to find out the optimal combinations through evolutionary computation. The experimental results have shown that the proposed approach is more efficient and effective in terms of user-defined ε values for different QoS objectives.