This paper provides a classification methodology of Malayalam characters segmented from scanned document images. Optical Character Recognition (OCR) is one of the successful area which has wide variety of applications related to pattern recognition. This paper describes segmented character recognition using Singular Value Decomposition (SVD). Euclidean distance measure is used for finding the nearest character class of the segmented character image during testing. For each character class, a resultant template is created from training character images using the proposed approach, which in turn reduces the comparisons required for classification. The result obtained from the experiment shows that this method provides an accuracy of 97%.