Light-emitting diode (LED) technology offers the possibility of obtaining white light, despite narrow-band spectra. In order to characterize the colour discrimination efficiency of various LED clusters, we designed a classification test, composed of 32 caps equally distributed along the hue circle at about 3 Delta E-ab*-unit intervals. Forty normal colour observers were screened under four different LED test light sources adjusted for best colour rendering. and under one control incandescent light of the same colour temperature. We used commercially available red, blue, and/or amber LED clusters. These yielded a poor colour rendering index (CRI). They also induced a significantly higher number of erroneous arrangements than did the control light. Errors are located around greenish-blue and purplish-red shades. parallel to the yellow-axis direction, whereas when the distribution of, light covers the full spectrum, the LED clusters achieve satisfactory colour discrimination efficiency With respect to the lights we tested, the colour discrimination is correlated with the CIE CRIs as well as with a CRI based on our sample colours. We stress Me fact that increasing the chroma of samples by lighting does not necessarily imply an improvement of colour discrimination. (c) 2008 Wiley Periodicals. Inc. Col Res Appl, 34, 8-17. 2009: Published online in Wiley InterScience DOI 10.1002/col.20459