Social psychology and neuroscience had confirmed that emotion state exerts a significant effect on human communication, perception, social behavior and decision making. With the wide availability of smartphones equipped with microphone, accelerometer, GPS, and other source of sensors, it is worthwhile to explore the possibility of automatic emotion detection via smartphone sensing. Particularly, we focus on a novel research problem that tries to detect the compound emotion (a set of multiple dimensional basic emotions) of smartphone users. We observe that users' self-reported emotional states have high correlation with their smartphone usage patterns and sensing data. Based on the observations, we exploit a feature extraction and selection algorithm to find the most significant features. We further adopt a factor graph model to tackle the correlations between features and emotion labels, and propose a machine learning algorithm for compound emotion detection based on the smartphone sensing data. The proposed mechanism is implemented as an APP called MoodExplorer in Android platform. Extensive experiments conducted on the smartphone data collected from 30 university students show that MoodExplorer can recognize users' compound emotions with 76.0% exact match on average.