The increasing availability of affordable color raster graphics displays has made it important to develop a better understanding of how color can be used effectively in an interactive environment. Most contemporary graphics displays offer a choice of some 16 million colors; the user's problem is to find the right color. Folklore has it that the RGB color space arising naturally from color display hardware is user-hostile and that other color models such as the HSV scheme are preferable. Until now there has been virtually no experimental evidence addressing this point. We describe a color matching experiment in which subjects used one of two tablet-based input techniques, interfaced through one of five color models, to interactively match target colors displayed on a CRT. The data collected show small but significant differences between models in the ability of subjects to match the five target colors used in this experiment. Subjects using the RGB color model matched quickly but inaccurately compared with those using the other models. The largest speed difference occurred during the early convergence phase of matching. Users of the HSV color model were the slowest in this experiment, both during the convergence phase and in total time to match, but were relatively accurate. There was less variation in performance during the second refinement phase of a match than during the convergence phase. Two-dimensional use of the tablet resulted in faster but less accurate performance than did strictly one-dimensional usage. Significant learning occurred for users of the Opponent, YIQ, LAB, and HSV color models, and not for users of the RGB color model.