I propose a strategy for the empirical evaluation of prospect theory that links concepts from the literature on asset pricing with heterogeneous agents to behavioral finance. I develop an asset pricing model in which two representative agents maximize their utility by investing in risky assets. One agent represents the behavior of investors above their reference level, one below. Using US income panel data, investors are sorted into groups depending on recent income development. In line with prospect theory, estimation results show that investors below their reference level act risk-seeking. The cross-sectional variation in returns of portfolios sorted by size and book-to-market value can be explained with a plausible risk aversion coefficient of ten while the unexplained equity premium is drastically reduced.