We propose a novel scheme for predicting the recognition performance by using confidence intervals. Biometrics is the best solution for preventing illegal sharing of authentication solutions (e.g., password, identity card) in a multi-factor authentication system. However, information acquired using sensors contains considerable noise, which can lead to errors. Our proposed scheme provides an efficient solution for treating noise and the resulting errors. When an image acquired using the sensor includes considerable noise and the estimated matching result is “fail,” the proposed scheme offers a short path. Our experimental results show the efficiency of the proposed scheme.