Abstract: A ear recognition method was proposed which combines the shape feature of outer ear contour with the structural feature of inner ear. The longest axis of auricle contour, which is the line crossing through the two points that have the longest distance on auricle contour, was firstly extracted after edge detection. Then least squares fitting polynomials were used to approximate the auricle curves. The coefficients of the polynomial functions were taken as the feature vector of outer ear contour. The longest axis intersects with the inner ear curves at several points. The ratio of the section length between the intersection points to the longest axis acts as the feature vector of inner ear. Owe to relevant invariability of the longest axis, the feature vectors extracted in the method are invariant to scale, parallel move and rotation. Experiment results prove robustness of this method.