The use of Raman spectroscopy for biomedical applications requires overcoming the obstacle of the broad background that is also generated by biological samples. This background, which is often largely attributed to fluorescence, is frequently orders of magnitude greater than the Raman signal and needs to be removed in order to use Raman spectra in sample analysis. Several methods have been proposed for removing fluorescent signal, both instrumental and computational. Of the computational methods, polynomial fitting has become increasingly popular. Typically, a polynomial of approximately fifth order is used in the fitting. This method alone is not always capable of fitting some more tightly featured spectra that may be present in data, potentially coming from a contaminant in the sample itself or from the experimental design. If this signal is present in varying amounts, the polynomial background removal method can leave the residual spectra with non-uniform artifacts that hinder classification results. If a reference spectrum can be obtained for this interfering signal, however, it can be incorporated into the polynomial fit and removed separately. An automated method for the removal of broad and/or moderately featured background signal is described. In addition to simulations, the method has been applied to spectra from biofilms of Streptococcus mutans.