Adaptation to a spatial or temporal context changes the color within that context. Since colors can be represented by their coordinates in a convex cone embedded in the three-dimensional real vector space, the influence of the context can be modeled by linear, affine or projective transformations. Because only projective transformations predict an invariance of the spectrum locus, the effect of a context surrounding a highly saturated stimulus could differentiate between these models. A new experimental setup based on a liquid crystal tunable imaging filter (LCTF) was developed to present a variable highly saturated or even monochromatic stimulus surrounded by a colored context. Four subjects produced cross context matches by adjusting the stimulus within the target context to look the same as a similar stimulus previously presented within another context. The resulting data can be described neither by a linear or an affine or projective transformation. The "pure" context-effect, adjusted for the effect of memory for colors, reveals a better fit of the projective model. There are further indices favoring the projective model: If the linear or affine models were valid, subjects should not be able to match certain highly saturated stimuli within the target context but in fact they do so. They even rate these matches especially well and produce them faster and more efficient than other matches. These results strongly suggest, that a change of the context leads to a projective transformation of the colors contained within that context.