This paper describes a feature extraction method for optical character recognition system for handwritten documents in Malayalam, a South Indian language. The scanned image is first passed through various preprocessing stages of operations like noise removal, binarization, thinning and cropping. After preprocessing projection profiles of each character is found. 1- D Discrete Cosine Transform (DCT) of projection profiles used as a feature. A multilayer artificial neural network (ANN) with logsig activation function is used for classification. The promising feature of the work is that successful classification of 44 handwritten characters.