Abstract. This paper illustrates how to design a visual experiment to measure color differences in gonioapparent materials and how to assess the merits of different advanced color-difference formulas trying to predict the results of such experiment. Successful color-difference formulas are necessary for industrial quality control and artificial color-vision applications. A colordifference formula must be accurate under a wide variety of experimental conditions including the use of challenging materials like, for example, gonioapparent samples. Improving the experimental design in a previous paper [Melgosa et al., Optics Express 22, 3458-3467 (2014)], we have tested 11 advanced color-difference formulas from visual assessments performed by a panel of 11 observers with normal color vision using a set of 56 nearly achromatic color pairs of automotive gonioapparent samples. Best predictions of our experimental results were found for the AUDI2000 color-difference formula, followed by color-difference formulas based on the color appearance model CIECAM02. Parameters in the original weighting function for lightness in the AUDI2000 formula were optimized obtaining small improvements. However, a power function from results provided by the AUDI2000 formula considerably improved results, producing values close to the inter-observer variability in our visual experiment. Additional research is required to obtain a modified AUDI2000 color-difference formula significantly better than the current one.