This paper deals with automatic generation of fundamental frequency (F0) contours for standard Arabic language. We use synthesis by rule for modelling Arabic intonation. The proposed model is based on the assumption that linguistic information is contained in target points of intonative contour. The perception of Arabic lexical accent is correlated with variations of F0. Rules used to determine target points are based particularly on accentuation algorithm. The existence of a declination phenomenon is discussed and introduced in our model. The validation of this model is carried out by TTS system based on TD-PSOLA technique. We provide an evaluation of our results based on perceptive test. The proposed processing of intonation allows a noticeable improvement of speech synthesis naturalness.