We have developed a novel noise-minimization method that is applicable to all kings of spectral data,which introduces little distortion in the signal waveform. It is based on the singular value decompositionof a matrix and is an advanced version of the conventional method for minimizing the background noisefrom a line spectrum, the details of which we reported in a previous paper [Appl. Spectrosc. 48, 1453(1994).]. In order to cope with a continuum spectrum, we have calculated a difference spectrum betweenan observed spectrum and a smoothed one. After minimizing noise components from the differencespectrum by the use of the previous method, we add it to the smoothed spectrum. We also propose amethod for reducing the computation time by introducing a data-division technique. Noise minimizationfor an infrared absorption spectrum of polystyrene is shown.