Reference-dependent preferences have been one of the most important contributions of behavioral economics. They have helped resolve empirical puzzles in many areas of economics: from explaining risk aversion over small stakes to the endowment effect to the failure of the predictions of models of intertemporal labor supply. In each case, a model in which losses relative to a reference outcome loom larger than gains can explain the observed anomaly. More, recently, the question of where this reference point may come from has arisen. A series of papers have developed models in which the reference point is shaped by expectations, and expectations are rational. These models have a clear theoretical appeal as they capture an important intuition present in most of the applications and lend themselves to elegant formalizations. In this talk, I give a brief overview of the existing literature and initial evidence that support the view that reference points are shaped by expectations. I then turn to some two new studies that test the predictions of the model in two different areas. In each of them, the model performs poorly and points to significant problems that reference points are formed by expectations. I conclude by outlining what the evidence suggests for modeling of reference points.