Effective data source selection strategy is one of the important issues of the spatial information retrieval. This paper proposes a method to select data source based on spatial scale semantics. The method determines the data source set for the distributed query plan fully considering the human vision features, filters those data sources which don't contribute to the final result. Through prefetching the simplified representation of data sources, the method can further reduce the overall cost of a query. The experiment shows that the proposed method has favorable performance both in small and large size query.