Most existing Web service search engines employ keyword search over databases, which computes the distance between the query and the Web services over a fixed set of features. Such an approach often results in incompleteness of search results. The Earth Mover's Distance (EMD) has been successfully used in multimedia databases due to its ability to capture the differences between two distributions. However, calculating EMD is computationally intensive. In this paper, we present a novel framework calledWS-Finder, which improves the existing keywordbased search techniques for Web services. In particular, we employ EMD for many-to-many partial matching between the contents of the query and the service attributes. We also develop a generalized minimization lower bound as a new EMD filter for partial matching. This new EMD filter is then combined to a k-NN algorithm for producing complete top-k search results. Furthermore, we theoretically and empirically show that WS-Finder is able to produce query answers effectively and efficiently.