Gradient is an important image processing tool allowing to carry out edge detection, segmentation, and texture analysis. It is expected to provide significantly more accurate and complete information when combined with the hyperspectral imaging technology. And such gain in accuracy justifies the cost and complexity of a hyperspectral acquisition. But how to measure accuracy? This article presents our first study in the accuracy assessment of hyperspectral gradient, where an assessment protocol based on a basic edge detection task is proposed. This protocol is then used to evaluate the full-band gradient approaches, where the results suggest to improve the protocol to include more complexity.