Curve-fitting procedures deal with an unknown, variable baseline by modeling it with a function involving a number of parameters In view of the facts that, (1) there is often no analytically relevant information in the baseline and (2) there is usually no functional form known, a priori for the baseline, we have chosen to eliminate it by means of the second derivative transformation. The resulting profile is deconvoluted by fitting it with the second derivative of the sum of an appropriate number of component curved. The utility of this procedure is demonstrated on simulated data with typical baselines and noise level, and is real FT-IR data. Peak parameters (such as position, width and area) detained from that technique are comparable to those obtained by fitting the original spectrum with Lorentzian curves and a simple baseline parameters that must be optimized in the fitting method. Applications of the technique could eliminate contributions from other complex baseline profiles in the quantitative analysis of spectral components.