Abstract: Targeted at the problem that traditional detection technology for Android malicious applications can no longer effectively detect malicious applications that are in explosive growth, which brought great challenge to the security of the mobile terminal, a classification algorithm of convolutional neural network (CNN) in deep learning was adopted to design and implement a detection solution for malicious applications based on static permission characteristics. First, the Android application package was decompiled to obtain the AndroidManifest.xml file, and the system permissions of applications were extracted. Second, the permission list based on the risk level of permissions were characterized, and the data set of permission characteristics was obtained. Furthermore, a convolutional neural network was trained repeatedly to obtain the application classifier. Finally, the classifier was used to determine whether the application contains malicious code or not. Results show that the accuracy of the detection solution can reach 98.8% and can efficiently identify malicious applications on Android platform to address security threats.