Nowadays, various apps greatly facilitate children’s lives and studies, while some apps also make illegal and inappropriate content (e.g., gambling, pornography) more accessible to children and adolescents. As the primary source of apps, several app markets adopt maturity ratings for apps, enabling users to distinguish whether apps are age-appropriate. However, if an incorrectly-rated app is acquired by users who are not of the appropriate age, it will bring severe consequences, especially for children. Giving an accurate maturity rating to an app can be time-consuming, both for developers and app market reviewers, while automatic rating tools can help solve this problem. Existing work on automatic app maturity ratings only analyzes app metadata obtained from app markets, but does not systematically consider the features of the apps themselves. In this work, we extract app features from both the app market and the apps themselves. We train machine learning models on Google Play, the official Android app market which has maturity ratings, and propose a cost-effective feature combination that achieves 96.98% accuracy, 96.21% precision, and 97.80% recall on within-market testing, and achieves 88.74% accuracy, 98.75% precision, and 83.72% recall on cross-market testing. Also, our method outperforms existing tools on every common metric.