In this paper, a novel approach based on the wavelet orthonormal decomposition is presented to extract features in pattern recognition. The proposed approach first reduces the dimensionality of a two-dimensional pattern, and thereafter performs wavelet transform on the derived one-dimensional pattern to generate a set of wavelet transform subpatterns, namely, several uncorrelated functions. Based on these functions, new features are readily computed to represent the original two-dimensional pattern. As an application, experiments were conducted using a set of printed characters with varying orientations and fonts. The results obtained from these experiments have consistently shown that the proposed feature vectors can yield an excellent classification rate in pattern recognition.