Among many alternative models developed in response to recent experimental evidence, the reference-dependent model emerged as one of the more realistic and more capable of capturing individual choices in the static situation. In this thesis, I explore whether reference-dependent model can also be applied to one of the most important human activity, intertemporal decision-making. In chapter one, I review three prominent economic theories. I also give a brief review of experimental studies on intertemporal decision-making. In chapter two, I extend the reference-dependent preference model to intertemporal choices. The model is based on the idea of loss aversion studied by Kahneman and Tversky (1979), and the idea of mental accounting studied by Thaler (1985, and 1999). Here, an individual’s utility depends not only upon her current consumption level but also her current reference level. Hypotheses of how an individual’s reference is formed are also explored. An individual’s reference could either be endogenously determined by option bundles, exogenously determined by her anticipations of outcomes, or a hybrid reference level. In chapter three, we develop a model of reference-dependent preferences, where an individual is uncertain about her future circumstances. Here, an individual’s utility is composed of an intrinsic consumption utility and a reference-dependent gain-loss utility, and her reference levels follow a random walk process. While the model does not explicitly include time discounting and no return on saving, it nevertheless may offer an alternative explanation of present bias and negative time preference (future bias). In chapter four, I explore whether subjects’ behaviour in a within-subject choice experiment is consistent with a number of existing and emerging theories, namely hyperbolic discounting theory, Rubinstein’s similarity procedure, and a novel model of reference-dependent preferences. The major results of the within-subject choice experiment are that none of these three theories explain subjects’ behaviour. However, among these three theories, the novel model of reference-dependent preference performs no worse than Rubinstein’s similarity procedure, and hyperbolic discounting theory performs the worst with subjects’ behaviour.