The condition monitoring and fault diagnosis of rolling element bearings play an important role in the safe and reliable operation of rotating machinery. Feature extraction based on vibration signals is an effective means to identify the operating condition of rolling element bearings. Methods based on multi-scale mathematical morphology (MM) have recently been developed to extract features from one-dimensional signals. In this paper, a new double-dot structuring element (SE) is constructed for multi-scale MM. A pattern spectrum, obtained from the multi-scale MM, is used as a feature extraction index. A correlation analysis gives the final identification result by utilizing information over a whole pattern spectrum. Compared with the most commonly used flat SE, the double-dot SE can extract more features of original signals at different scales. Vibration signals, measured from defective bearings with outer race faults, inner race faults and ball faults, are used to evaluate the fault detection ability of the proposed SE and bearing fault diagnosis method. Results show that faults at different levels can be identified, including ball fault; and the location of outer race fault can also be differentiated.