In recent years, the number of malware on the Android platform has a geometrical growth. Therefore, it is very necessary to have a method to detect Android malware. This study experiments with a large number of Android malware samples and machine learning technology to establish a prediction model for malware classification, which is run in the static detection process. First, we obtain the permissions and the dangerous API information of Android applications, the permissions feature in its AndroidManifest.xml file by decompiling APK files and its dangerous API features by translating decompiles class.dex files into smali files together with the baksmali tool. Then, we use multiple classification algorithms and preprocessing algorithm to compare the accuracy rate of the single detection and the conjoint detection. The experimental results show that the accuracy rate of the conjoint detection is higher than that of the single detection, and the accuracy rate reaches up to 97.5%.