Psychophysical experiments were carried out to collect the observer accuracy data which are used to analyse the performance of state of the art colour difference formulae, and to derive a colour difference model for moving images. Three MPEG standard test streams were used to the study. The initial hypothesis for the data analysis of moving images is that the moving images are combinations of consecutive still images. Seven image quality attributes were asked to the observers. colour difference thresholds were analysed before the calculation of colour differences. The result showed that human visual system (HVS) is highly sensitive to the difference of memory colours especially to skin tone, and the sensitivity decrease along with the increase of spatial and temporal frequencies. Then the performance analysis of state of the art colour difference formulae such as CIELAB, CIE94, CIELAB CMC, CIEDE2000, SCIELAB and iCAM were performed to see which formula can best predict the differences of various image quality attributes for moving images. Wrong decision (WD) analysis was used and the results showed that CIELAB performed best for question of overall image difference. For sharpness difference, SCIELAB performed best followed by CIELAB. Colour difference models for moving images are proposed by use of temporal blur filter to the array of corresponding pixels throughout the sequences of images. CIELAB colour difference model with temporal blur performed best and it can be the final candidate of the novel colour difference model for moving images.