Fluorescence staining is the prevailing image acquisition technique in cell biology research. A problem of this method is the incidental fluorescence, that is the total fluorescence collected by a camera pixel originating from all sources in the sample except for the targeted point. As a result, typical images exhibit blurry regions at the periphery of fluorescence stained cell structures. Generic image processing algorithms address the problem indirectly and with limited success. We propose a technique directly targeting incidental fluorescence by optimizing a simple physical model. Our results indicate that incidental fluorescence can reliably be reduced, and often eliminated, without a significant loss in content. Resulting images are considerably better for human vision, while improvements are actually realized in automated processing. We demonstrate the performance of the technique on microtubule images that are known to suffer from incidental fluorescence.