To deal with the localization problem of robot equipped with monocular camera, a novel Monte Carlo method based on Scale Invariant Feature Transform (SIFT) is proposed. The features are extracted by modified SIFT to make the features be invariant to changes in illumination, scale, 3D viewpoint and noise, and to reduce the number of features generated by SIFT as well as their extraction and matching time. During robot motion, the information from feature observations is fused with that from the odometry by particle filter, which improves the speed and accuracy of the localization effectively. Experimental results demonstrate the validity of the approach.