Character skeleton plays a significant role in character recognition. This paper presents a novel algorithm based on multiscale approach to extract skeletons of printed and hand-written characters. The development of the method is inspired by some desirable characteristics of the modulus minima of wavelet transform. Namely, the local minima of wavelet transform are scale-independent and locate at the medial axis of the symmetrical contours of character stroke. Thus it is particularly suitable for characterizing the inherent skeletons of character strokes. The proposed skeletonization algorithm contains two major steps. First, by thresholding for the modulus minima of wavelet transform, the modulus minima points underlying the character strokes are extracted as the primary skeletons. Based on these primary skeletons, the modulus minima points are being eventually computed as the final skeleton by iteratively performing wavelet transform. The skeleton form the proposed method can be exactly located on the central line of the stroke, and the artifacts and branches of skeletons from traditional methods can be avoided. We tested the algorithm on handwritten and printed character images. Experimental results indicate that the proposed algorithm is applicable to not only binary image but also gray-level image.