In this work, we focus on an improvement of a multi-script handwritting recognition system using a HMM based classifiers combination. The improvement relies on the use of Dempster-Shafer theory to combine in a finer way the probabilistic outputs of the HMM classifiers. The experiments are conducted on two public databases written on two different scripts : IFN/ENIT (latin script) and RIMES (arabic script). The obtained results are compared with the classical algorithms of the field and the superiority of the proposed approach is shown.