The effect of image processing in computed radiography (CR) has been analyzed in many ROC studies. The results have not shown great diagnostic improvements, except in some special occasions. The theoretical effect of image enhancement on the signal-to-noise ratio in CR images has so far not been assessed. Concerning the previous results, the changes induced in the signal-to-noise ratio by digital image processing are certainly of interest. We calculated the signal-to-noise ratio in various conditions according to the principles of the Rose model, using the computerized image data of storage phosphor radiography. Seventy-seven computed radiographs processed by Gaussian unsharp-mask filtering using different kernel widths were analyzed. The signal-to-noise ratio was reduced in all images by more than 40% when the smallest kernels were used, and increased slowly towards the original value with greater kernel sizes. In no conditions did the ratio exceed the original one. The results show that although edges and signal contrast can be enhanced by unsharp-mask filtering, this happens at the cost of increased noise. This might at least in part explain why image processing does not significantly improve the diagnostic information content of a computed radiograph.