We are witnessing a large increase in satellite generated data specially in the form of images. Hence intelligent processing of the huge amount of data received by dozens of earth observing satellites, with specific satellite image oriented approaches, presents itself as a pressing need. Content based satellite image retrieval (CBSIR) approaches have mainly been driven so far by approaches dealing with traditional images. In this paper we introduce a novel approach that refines image retrieval process using the unique properties to satellite images. Our approach uses a Query by polygon (QBP) paradigm for the content of interest instead of using the more conventional rectangular query by image approach. First, we extract features from the satellite images using multiple tiling sizes. Accordingly the system uses these multilevel features within a multilevel retrieval system that refines the retrieval process. Our multilevel refinement approach has been experimentally validated against the conventional one yielding enhanced precision and recall rates.