Grid infrastructure is becoming a platform for the design, implementation and execution of various scientific experiments. Consequently it forms a hidden storehouse of application knowledge about historically how a grid application was designed and performed (e.g. the used approaches, required data, etc). This work addresses the issue of abstracting the application knowledge in grid based on service provenances. Our method is composed by the steps of providing a formal description for detailed service provenances; deducing the set of service provenance descriptions that indicates how a desired application result can be obtained; abstracting and visualizing the deduced result to facilitate the understanding, sharing and reuse of corresponding application knowledge. The implementation status of our method is also reported.