Sleep is an essential phase in human circadian rhythm with importance in restoring human vigour and vitality. Conventional sleep examination is done using polysomnography with many sensors connected to various parts of human body. Recently, research in sleep is geared toward alternative feasibility of using only ECG signal. In this research, sleep stages classification using only features derived from single-lead ECG is conducted. The capability of generalized learning vector quantization (GLVQ) in discriminating sleep stages is tested. We showed that GLVQ which is configured with multi codebooks shows promising result in discriminating complex sleep stages data.