Transcription factor binding sites (TFBS) are often located in the upstream regions of genes and transcription factors (TFs) cause transcription regulation by binding at these locations. Predicting these binding sites is a difficult problem, and traditional methods have a high degree of false positives in their predictions. Comparative genomics data can help to improve motif predictions. In this paper, a new strategy is presented, which refines motif by taking the comparative genomics data into account. Tested with the help of both simulation data and biological data, we show that our method makes improved predictions. We also propose a new metric to score a motif profile. This score is biologically motivated and helps the algorithm in its predictions.