Abstract: This paper proposes a new method for automatic building recognition and mapping in remote sensing images. The method calculates the multiple feature of building first, and then selects the best feature list with PCA algorithm to segment the imagery into two kind of area, which contain building targets and does not contain. Finally, a method combined both the dominant direction and area factor is used to reconstruct the target buildings into regular polygons. The experimental results prove that this new method has a good precision and robustness.