Health condition identification of planetary gearboxes is crucial to reduce the downtimeand maximize productivity. This paper aims to develop a novel fault diagnosis methodbased on modified multi-scale symbolic dynamic entropy (MMSDE) and minimum redun-dancy maximum relevance (mRMR) to identify the different health conditions of planetarygearbox. MMSDE is proposed to quantify the regularity of time series, which can assess thedynamical characteristics over a range of scales. MMSDE has obvious advantages in thedetection of dynamical changes and computation efficiency. Then, the mRMR approachis introduced to refine the fault features. Lastly, the obtained new features are fed intothe least square support vector machine (LSSVM) to complete the fault pattern identifica-tion. The proposed method is numerically and experimentally demonstrated to be able torecognize the different fault types of planetary gearboxes.