Suppose that we have a set of discrete data of a noisy signal and we reconstruct the original signal from its orthogonal moments based on Legendre polynomials. In this paper we derive a new error estimation of the reconstruction, which reveals the relationships between the number of moments, the signal smoothness and sampling rate. Numerical simulations provided are consistent with our analysis. (C) 2002 Elsevier Science Inc. All rights reserved.13241700249263