The study of multiple classifier systems has become recently an area of intensive research in pattern recognition. Also in handwriting recognition, systems combining several classifiers have been investigated. An approach for recognizing the legal amount for handwritten Arabic bank check is described in this article. The solution uses multiple information sources to recognize words. The recognition step is preformed with a parallel combination of three kinds of classifiers using holistic word structural features. The classification stage results are first normalized, and the sum combination is performed as a decision fusion scheme, after which a syntactic analyzer makes final decision on the candidate words. Using this approach, the obtained results are very interesting and promising.