The visual sensation of gloss is built on cues deduced from the interaction between light, surfaces sunder evaluation, and surrounding conditions. Gloss is a second-order attribute of the visual appearance, this means that its perception is not directly encoded on biological sensors but constructed from the global scene in the field of view of the observer. It is then a complex quantity to measure. When most studies based on simulated samples stress on the importance of realistic observation conditions, we measure the effect of environment complexity over perception of real samples. We test two different lighting conditions: either diffuse or a combination of diffuse and collimated lighting in order to approach natural complex illumination patterns. Under both lighting conditions, we test two environments: a standard black light booth, designed according to the ASTM D4449, and a realistic office cubicle. Samples consist in a seven-level gloss scale ranging from full matt to high gloss. These are presented to observers through pair comparison protocol, according to a maximum likelihood difference scaling algorithm. Our results confirm that gloss constancy is maintained even if the convergence of illumination varies. We however measure that the constancy is lost for matt samples perception.