The interest for spectral color reproduction has increased with the growing field of multispectral imaging and the increasing use of multi-colorant printing systems. Spectral color reproduction, i.e. aiming at reproducing the spectral reflectance of an original, first requires a colorant separation for a multi-colorant printing system, followed by halftoning of each the color channels. Spectral vector error diffusion, sVED, has previously been introduced as a tempting alternative for spectral color reproduction, since the method combines the colorant separation and the halftoning in a single step. Only the spectral properties of the Neugebauer primaries are needed as input, and there is no need to invert a complex printer model for the colorant separation. Previously, spectral vector error diffusion has been positively evaluated for simulated prints, assuming a perfect printer and no dot gain. In this study, we evaluate the performance of sVED in practice, for real prints.Spectral vector error diffusion has been used to reproduce 1000 spectral targets, all within the spectral gamut of the printing system. The resulting color patches have been printed in various print resolutions, using a 10-colorant inkjet printing system. The experimental results reveal a remarkably large difference between the reproduction errors for the printed samples compared to the simulated spectra from the digital halftones. The results show a strong relation between the print resolution and the magnitude of the reproduction error, with lower resolutions giving smaller errors, due to the effect of dot gain in the printing process. The experimental results imply that in its current form, without compensation for physical and optical dot gain, spectral vector error diffusion produces unacceptable spectral and colorimetric reproduction errors, for any print resolutions used in practice.The results further show that the sVED method in many cases produces color patches that appear noisy and visually unpleasant. By replacing the spectral RMS difference with the ΔE₉₄ color difference as criterion in the sVED algorithm, the graininess as well as the resulting color difference was decreased. However, the improvements in colorimetric performance and more visually pleasant reproductions, comes at the cost of an increase in spectral reproduction errors.