Abstract. Anti-spoofing protection of biometric systems is always a serious issue in real-life applications of an automatic personal verification system. Despite the fact that face image is the most common way of identifying persons and one of the most popular modalities in automatic biometric authentication, little attention has been given to the spoof resistance of face verification algorithms. In this paper, we discuss how a system based on DCT features with a likelihood-ratio-based classifier can be easily spoofed by adding white Gaussian noise to the test image. We propose a strategy to address this problem by measuring the quality of the test image and of the extracted features before making a verification decision.