The human facial profile curve provides complementary information of the face that is not present in the frontal-view face, which has been used in face identification, face analysis and modelling. This paper addresses robust facial profile recognition. With appropriate rotation, the profile curve can be considered as a histogram, where histogram metric could be employed to measure profiles. The advantage is that no detection of fiducial points is required, which is usually unreliable and hard to implement fully automatically. This paper also introduces three methods to align profiles, and investigates four similarity measures. The experiments on two profile image databases (Bern and FERET) and a facial range data set are carried out. The comparison with two primary approaches is conducted. The experimental results demonstrate that, compared with other methods, the involved metric for profile recognition has perfect performance robustness against noise.