Optical Character Recognition techniques for printed and handwritten text are quite different. Therefore, before any further document preprocessing, it is necessary to separate these text types. A fundamental step for this separation is the segmentation. In this paper we address the problem of segmentation these documents into words. The proposed system was tested in two public image databases. Many measures of efficiency were computed achieving correct separation results above 96% with respect to mean precisions and 97% for average of the accuracies. Although it would be very important compare our results with some other algorithm for the same purpose, on this moment it is impossible because there is no work in the same purpose were such comparison could be done.