Optimal sensitivity of fluorescence in situ hybridization (FISH) requires bright signals and low background fluorescence. Use of locus-specific probes is especially dependent on high sensitivity. Some tissue preparations show high autofluorescence, masking small or dim signals. We have developed a new method for subtracting autofluorescence from digital images on a pixel-by-pixel basis. It is based on the observation that fluorescent labels for FISH have narrower excitation and emission spectra than the chemical components responsible for autofluorescence. Our new approach uses calculation of the ratio of autofluorescence between multiple color images for correction of autofluorescence in each individual image. By subtracting autofluorescence components, we were able to enhance centromeric signals and make previously indistinguishable cosmid signals clearly visible. This image-processing approach to autofluorescence correction may widen the applicability of gene-specific probes in FISH analysis of tumor material.