Median filtering of digital imagery is known to remove noise while preserving edges. However, it is not evident that median filters facilitate further machine processing such as machine recognition. In this paper, median filtering and median/inverse filtering are explored and evaluated based on their performances in machine recognition. Furthermore, this work is also aimed at gaining insight into the design of practical median-filter preprocessors for automatic pattern recognition systems. The design of a median filter depends solely on the selection of window shape and size. For this study, the median-filter window shape and size were determined empirically based on the performance in enhancing automatic pattern recognition. The results show that median filtering is most effective for enhancing machine recognition in cases in which images are corrupted by impulse-type noise alone or in conjunction with blurs. It is not as effective when images are corrupted by white Gaussian noise and/or in combination with blurs; however, median/inverse filtering does enhance machine recognition in these cases.