In off-line character recognition, reliable extraction of strokes greatly influences the overall performance of the system. We propose a novel method of decomposing a Chinese character into a set of strokes, based on the concepts of mathematical morphology. A character is segmented after being thinned using morphological operation. Each segment is expanded by morphological operations such as elongation, fattening and isotropic expansion then intersected with the original input shape. The expanded segments are merged using a convexity measure at the final stage. The final merged segments are regarded as basic strokes of a character. To show the effectiveness of the method, we perform experiments and the results show the validity of the proposed method.