Visual and instrumental colour matches between broad band lights and lights composed of the additive mixture of R, G, and B LEDs differ considerably. Previously we have shown that using the primaries suggested by Vos or those now under consideration by the CIE TC 1-36 will diminish the differences between visual and instrumental matches.In a new experiment also the question of the spectral power distribution of the RGB LED primaries has been investigated. We tested LEDs for which the dominant wavelengths were near to the prime colours suggested by Thornton, and as far as possible from these chromaticities, thus near to Thornton's anti-prime colours.Characteristic differences were found for the two sets of primaries, showing that colorimetric settings using the anti-prime colours does not provide acceptable colorimetric description. This has been shown by colouring images in the following way: A hyper-spectral image was taken, where the reflection spectrum was known for every pixel, and in a simulation this image was irradiated by two RGB LED lights of equal correlated colour temperature, one using the prime colours, the other using the anti-prime colours. Comparing the two simulations showed far larger differences then could be expected from interaction of the broad band reflection spectra and the narrow band LED spectra.