Readability plays an important role in the process of choosing materials adapted to readers' levels. Indeed, the task of automatically measuring the readability of a text frequently faces challenges such as lack of data and resources.In this thesis, we have produced a set of models for automatic readability prediction. These models are dedicated to readers of Arabic as a first language (L1) and as a foreign language (L2). The development of these models involves three main stages. The first step consists in collecting a reference corpus annotated according to levels of difficulty. The second step consists in converting the texts into feature vectors. The final stage in the development of a predictive readability model is a classification phase.In the final part of this thesis, we have exploited all the resources and models we have developed for an initial investigation into the field of text simplification.