Accurate detection of facial feature points is crucial to analyze facial features and to synthesize caricatures by computer. There have been many works on the automatic detection of facial feature points. However the accurate detection methods which can be used in above-mentioned applications have not been developed yet. This paper proposes the method to improve accuracy in the detection of facial feature points using the statistical characteristics of their locations. Statistical characteristics are examined by using many face images. They are used both to check the misdetection of feature points and to locate them at the appropriate locations. Experimental results using 247 face images of males and females from various ages show that the proposed method improves the accuracy in detection of feature points for eyebrows, mouth and facial contour.