Loss aversion, the fact that losses have a greater impact than gains, is a fundamental property of behavioral accounts of choice. In this paper, we suggest four possible characterizations of the relative impact of losses and gains: (1) It could be a constant, such as the much cited value of 2, as in losses have twice the impact of gains. (2) It could be a systematic individual difference, with some individuals more or less loss aversion, (3) it could be a property of the attribute, or (4) a property of the different processes used to construct selling and buying prices. We examine the behavior of a large sample of auto buyers using an experiment which allows us to measure loss aversion, at the individual level for several different attributes. A set of hierarchical linear models shows that to understand loss aversion, one must consider the process used to construct prices. Interestingly, we show that knowledge of the attribute lowers loss aversion and that age and attribute importance increases loss aversion.