An improved predictor of colourfulness in the Hunt 91 model of colour vision is described. A new chroma predictor allows for the effect of the luminance Jactor of the background more effectively, and results in more consistent predictions for different viewing conditions. The new colourfulness predictor is obtained by multiplying the new chroma predictor by a factor that varies more slowly with the level of adapting luminance than was the case for the old colourfulness predictor; this results in much better predictions, whose coefficients of variation are similar to those that characterize observer consistency. The steps required for reversing the model are given in an appendix.