Segmentation of cursive text has been one of the major problems in Arabic writing. The problem is the shape of the letter which is context sensitive, depending on it s location within a word. Many text recognition systems recognize text imagery at the character level and assemble words from the recognized characters.Unfortunately this approach does not work with Arabic text. In this paper we describe a new approach to segment Arabic text imagery at a word level, without analyzing individual characters. This approach avoids the problem of individual characters segmentation, and can overcome local errors in character recognition.