This paper refers to the use of a modified set of moments for the development of efficient shape discrimination and classification techniques. These moments may be applied in general for the discrimination of other patterns as well, such as geological, medical and robotic vision signals. A new set of efficient moments is introduced, which is invariant under rotation, translation, and scale of the image. These moments are normalized with respect to standard deviation, they appear to have better classification performance over the existing sets of moments, and they are less sensitive to noise. Application of the proposed technique to the classification of silhouettes of airplanes is given. The discrimination of an unknown shape is achieved by testing a weighted least square function.