This paper describes an Arabic Synthesis-by-Rule System which is completely simulated on a PDP 11/34 minicomputer and produces an intelligible speech. The input to the system could be any typed Arabic text. The orthographic phonetic translation is designed as a set of arborescent tests relative to the right and left contexts of an analysis window that slides along the sentence. An automatic process of syllabification permits the localisation of the lexical stress. The duration of the vowels in determined by a set of timing rules. The phonetic module computes the acoustic parameters of each phonetic segment which are used to control a formant synthesizer. Enphasis is made on the processing of linguistic peculiarities of Arabic language.