Automatic modulation classification has become an important issue in various application areas such as civil and military domains. Fractal theory has been widely used for signal analysis and can provide effectiveness features such as fractal dimension for automatic modulation classification. However, most of the signals are not critical self-similar fractals; the traditional single fractal dimension can’t reveal the fractal characteristics of modulation signals. In this work, we explore the capacity of the multi-scale fractal dimensions to represent the complexity of modulation signals. The morphological covering (MC) method is selected to calculate the multi-scale fractal dimensions. Four typical modulation signals, mean the ASK, PSK, FSK and QAM signals, are simulated to evaluate the effectiveness of the presented method. Experimental results demonstrated that the multi-scale fractal dimensions calculated by morphological covering can satisfactorily distinguish the four modulation types. They can be further utilized as features for automatic modulation classification tasks.