This paper presents a fault diagnosis scheme based on morphological multi-fractal （MMF） analysis and improved grey relational analysis （IGRA） for rolling element bearings. In this scheme， firstly， the multi-fractal characteristics of bearing signals are illustrated by quality index and partition function. Secondly， the parameters of generalized fractal dimension and multi-fractal spectrum in different bearing operating conditions are calculated by morphology， from which some parameters with good discrimination ability are selected as fault-related feature. Thirdly， maximizing deviation is employed to improve the reliability of the classical grey relational analysis. Finally， the effectiveness of this method is verified by simulation analysis and application example. The results show that the proposed scheme can recognize the different fault categories， which is more stable and higher accurate than the traditional method， and the operation time is shorter， which is suitable for solving practical engineering problems.