The paper proposes a method of off-line handwritten digit recognition based on principal curves.The method uses principal curves and reduction of knowledge to extract the structural features of digits and design a classifier.Principal curves are nonlinear generalizations of principal component analysis.They are smooth self-consistent curves that pass through the "middle" of the distribution.They preferably reflect the structural features of the data.Reduction of knowledge is the efficient tool of obtaining classification rules from a decision table.Firstly principal curves are used to extract the structural features of training data.Secondly the classification features are chosen by analyzing the structural features of principal curves in detail

then we set up the decision table that consists of these classification features.Finally we automatically attain classification rules by attribute and attribute value reduction.The method accords with the recognition habit of human and overcomes the disadvantage of statistical features.The experimental result indicates that the method can effectively improve the recognition rate of off-line handwritten digits

and provides a new approach to the research for off-line handwritten digit recognition.