As the empirical mode decompositon（EMD） can accurately recognize the structure of the original signal,this paper proposes a new feature extraction algorithm of handwritten Arabic numerals based on wavelet transform and EMD.In this algorithm,first,smooth contours of numeral image are obtained by preprocessing the maximum module of the G wavelet transform.Then,an EMD analysis is performed to decompose the normalized curvature sequences into their components,which produces more compact curvature features.Finally,the obtained curvature features are clustered and recognized.Experimental results show that the proposed algorithm is superior to the classic feature extraction algorithm in terms of clustering effect and classifier design ability.