Accurate skew angle estimation is an essential component in document analysis system to enhance the performance of the optical character recognition (OCR). In this paper, a new and efficient moments based method to estimate skew angle of a pixels in the word in the scanned document image is proposed. The proposed technique has two stages. In the first stage, using boundary-growing method, pixels in the words of skewed text are extracted. The pixels in the words extracted are given as input to moments based method. It results in a skew angle in the second stage. Extensive experiments have been conducted on various types of documents such as documents containing different languages and different fonts to reveal the robustness of the proposed method. Comparative studies with the well-known methods are presented to show that the proposed method is superior in terms of accuracy and computational efficiency.