We describe a new, robust method of numerically extracting equivalent Raman spectra from coherent anti-Stokes Raman scattering (CARS) spectra. The extraction procedure requires no a priori information regarding the vibrational resonances or the shape of the nonresonant background (NRB). The method we present here is robust to situations where the NRB is not spectrally flat, and is accurate over a wide range of signal-to-background ratios. This simple and quick computation method, which relies primarily on a small number of fast Fourier transforms, could reasonably be used in conjunction with spectral imaging, where millions of spectra must be analyzed. Published 2009 by John Wiley & Sons, Ltd.