Three advanced CIELAB-based color-difference formulae, CMC, CIE94, and CIEDE2000, together with the basic CIELAB equation, were tested using large color-difference visual data (maximum average size was 12 CIELAB AE units) produced in this study. The color-difference comparison experiment was carried out at CIE Gray and Blue centers by a panel of 6 normal color-vision observers using CRT-generated stimuli based on the psychophysical method of constant stimuli. The experimental data, processed via probit analysis, were well fitted to chromaticity ellipses with a high reliance according to the observer accuracy in terms of PF/3 measure. A detailed comparison was performed to analyze the agreement between predicted color differences from all formulae and their corresponding visual scales in all measurement planes of CIELAB space. The results show that the CIEDE2000 marginally outperformed the others at all color centers while CIE94 was the worst in original formulae or with optimized kL value, but the CIELAB performed worst when the parametric factors of kL, kc, and kH were all optimized, with the CMC always lying between these extremes.