In this paper, we propose an analytical approach of an offline recognition of handwritten Arabic. Our method is based on Hidden Markov Models (HMM) Toolkit (HTK), modeling type that takes into consideration the characteristics of Arabic script and possible inclinations of cursive words. The objective is to propose a methodology for rapid implementation of our approach. To this end, a preprocessing phase that can prepare the data was introduced. These data are then used by an extraction method of two groups of the characteristics (Features of Local Densities and Features Statistical) with the use of the technique of sliding windows, the results of this step are processed in sequence information as vectors to HTK (Hidden Markov Model Toolkit).