As mobile payment has become increasingly popular recently, some phishing malwares on mobile devices try to steal users' sensitive information through forging a user interface similar to the targeted application, which cannot be detected by existing anti-phishing techniques. We propose a new similarity-based anti-phishing approach to detect phishing malware on Android. With the perceptual image hash technology, it compares the similarity of captured screen images between the suspicious application and the targeted application. If the similarity exceeds a threshold, the suspicious application is determined to be malicious. The experiment proves that the approach is effective to detect phishing malware on Android.