The colour resolution of a 14-bit and an 8-bit per channel graphics card were evaluated and compared with the just noticeable difference between colours (varying only in luminance) for: (1) a standard observer (based on the CIE 1976 L*u*v* colour space) and (2) real observers in a colour discrimination task. The results of this study show that an 8-bit per channel graphics card seems adequate for colour discrimination experiments where stimuli only vary in luminance. However, considering that the resolution of the graphics card should be equal to the Nyquist rate, an 8-bit per channel card turns out to be inadequate. For colour discrimination experiments where stimuli only vary in chromaticity, there is an undersampling of the colour space with respect to MacAdam ellipses when using 8-bit per channel graphics cards. The extremely fine colour resolution of a 14-bit per channel graphics card overcomes these problems. Its use allows more accurate measurements of achromatic and chromatic discrimination thresholds and avoids experimental (spatial or luminance) artefacts, such as bandings that can occur on achromatic or chromatic gradients. (c) 2010 Wiley Periodicals, Inc. Col Res Appl, 36, 118-126, 2011; Published online 21 May 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/col.20589