A subject tracking apparatus extracts a subject region which is similar to a reference image on the basis of a degree of correlation with the reference image for tracking a predetermined subject from images supplied in a time series manner. Further, the subject tracking apparatus detects the position of the predetermined subject in the subject region on the basis of the distribution of characteristic pixels representing the predetermined subject contained in the subject region, and corrects the subject region so as to reduce a shift in position of the predetermined subject in the subject region. Moreover, the corrected subject region is taken as the result of tracking the predetermined subject, and the reference image is updated with the corrected subject region as the reference image to be used for the next supplied image.