This paper presents a novel approach for word segmentation from handwritten Bengali document images. We have employed a modified version of the Scale Space method, where it is combined with the Shuffled Frog-Leaping algorithm (SFLA). Our approach overcomes certain limitations of the standard Scale Space method in choosing the right scaling parameters. The incorporation of the SFLA with the Scale-Space method for handwritten word segmentation allows adaptive parameter tuning and, in this way, optimizes the right scale for the segmentation process. This method is employed to segment words from handwritten Bengali answer sheets gathered from schools to develop a handwritten character recognition (HCR) system. The proposed method is compared with a few existing methods, and the experimental results show that the proposed method is superior to others.