I develop a model of endogenous bounded rationality due to search costs, arising implicitly from the decision problem's complexity. The decision maker is not required to know the entire structure of the problem when making choices. She can think ahead, through costly search, to reveal more of its details. However, the costs of search are not assumed exogenously; they are inferred from revealed preferences through choices. Thus, bounded rationality and its extent emerge endogenously: as problems become simpler or as the benefits of deeper search become larger relative to its costs, the choices more closely resemble those of a rational agent. For a fixed decision problem, the costs of search will vary across agents. For a given decision maker, they will vary across problems. The model explains, therefore, why the disparity, between observed choices and those prescribed under rationality, varies across agents and problems. It also suggests, under reasonable assumptions, an identifying prediction: a relation between the benefits of deeper search and the depth of the search. In decision problems with structure that allows the optimal foresight of search to be revealed from choices of plans of action, the relation can be tested on any agent-problem pair, rendering the model falsifiable. Moreover, the relation can be estimated allowing the model to make predictions with respect to how, in a given problem, changes in the terminal payoffs affect the depth of search and, consequently, choices. My approach provides a common framework for depicting the underlying limitations that force departures from rationality in different and unrelated decision-making situations. I show that it is consistent with violations of timing-independence in temporal framing problems, dynamic inconsistency and diversification bias in sequential versus simultaneous choice problems, and with plausible but contrasting risk attitudes across small- and large-stakes gambles.