An on-line recognition approach for the handwritten mathematical symbols is presented. Firstly, it analyses the structures of 94 commonly used mathematical symbols and concludes that all of them consist of 10 basic elements. Secondly, it proposes a new method of basic element ordering and reduces the number of standard symbols by extracting three primary features of mathematical symbols,namely,basic element vector, relative positions between basic elements,and basic element length vector. Finally, the traditional dynamic programming method is improved by considering matching and unmatching value and adding geometric restraints.Through improved Kohn-Munkres algorithm. During the recognition test of mathematical symbols handwritten by 20 subjects, correctness rate reaches 92.52%,incorrectness rate 3.03% and refusal rate 4.45%.