In This paper we presented new approach for cursive Arabic text recognition system. The objective is to propose methodology analytical offline recognition of handwritten Arabic for rapid implementation. The first part in the writing recognition system is the preprocessing phase is the preprocessing phase to prepare the data was introduces and extracts a set of simple statistical features by two methods : from a window which is sliding long that text line the right to left and the approach VH2D (consists in projecting every character on the abscissa, on the ordinate and the diagonals 45{\deg} and 135{\deg}) . It then injects the resulting feature vectors to Hidden Markov Model (HMM) and combined the two HMM by multi-stream approach.