Bounded rationality is the study of how human cognition with limited capacity is adapted to handle the complex information structures in the environment. This thesis argues that in order to understand the bounded rationality of decision processes, it is necessary to develop decision theories that are computational process models based upon basic cognitive and perceptual mechanisms. The main goal of this thesis is to show that models of perceptual categorization based on the storage of exemplars and retrieval of similar exemplars whenever a new object is encountered (D. L. Medin & M. M. Schaffer, 1978), can be an important contribution to theories of decision making. Study I proposed, PROBEX (PROBabilities from Exemplars), a model for inferences from generic knowledge. It is a “lazy” algorithm that presumes no pre-computed abstractions. In a computer simulation it was found to be a powerful decision strategy, and it was possible to fit the model to human data in a psychologically plausible way. Study II was a theoretical investigation that found that PROBEX was very robust in conditions where the decision maker has very little information, and that it worked well even under the worst circumstances. Study III empirically tested if humans can learn to use exemplar based or one reason decision making strategies (G. Gigerenzer, P. Todd, & the ABC Research Group, 1999) where it is appropriate in a two-alternative choice task. Experiment 1 used cue structure and presentation format as independent variables, and participants easily used one reason strategies if the decision task presented the information as normal text. The participants were only able to use exemplars if they were presented as short strings of letters. Experiment 2 failed to accelerate learning of exemplar use during the decision phase, by prior exposure to exemplars in a similar task. In conclusion, this thesis supports that there are at least two modes of decision making, which are boundedly rational if they are used in the appropriate context. Exemplar strategies may, contrary to study II, only be used late in learning, and the conditions for learning need to be investigated further.