A new simple algorithm, based on dynamic programming is presented, for handwritten character recognition, with improved accuracy. The proposed Shape Transform (ST) approach is based on the calculation of the cost of transforming the image of a given character into that of another, thus taking into account local geometrical similarities and differences. A large experiment is conducted on the NIST database. [2], and the effectiveness of the proposed method, is compared to the Karhunen Loeve Transform (KLT) method, with which a similar experiment was contacted [1] reporting the best results in the literature. The experiments performed, show that this new approach leads to improved recognition. It is more demanding in computer time, which is becoming ever more plentiful, but it lends itself to very efficient parallel hardware implementation.