We have developed a new method of quantitatively evaluating the gradation smoothness of output media. In this method, virtual smooth curves are first estimated from real gradation curves, which are outputted by a real device with a digital gradation chart. Roughness components are then extracted by subtracting the estimated smooth curves from the actual gradation curves in a uniform color space. Lastly, roughness corresponding to human perception is evaluated with an estimating function based on the roughness components. We conducted visual experiments to optimize the estimating function using simulated prints with a silver-halide photographic printer. We obtained a correlation coefficient of more than 0.80 between the subjective values and the index by the estimating function. We confirmed its performance with print samples outputted by six electro-photographic printers currently on the market. We further implemented this evaluation method as an Adobe Photoshop plug-in module for easy use.