This paper presents a comparison of two models, built on the same architecture, ACT-R, and on the same dynamic decision making task, RADAR. The two models represent the Strategy-Based Learning (SBL) approach and the Instance-Based Learning (IBL) approach. The SBL approach assumes a certain set of predefined strategies, and learning occurs by selecting the most successful strategy over time. The IBL approach proposes that decisions are made based on retrieval of good past experiences stored in memory. This approach assumes no previous initial experience apart from that gained while performing the task. Both models were tested with respect to two criteria: fit to human data during a training exercise with RADAR and adaptability to test conditions that are either similar to or different from the training conditions. Our comparison results demonstrate that both models fit learning human data successfully, but the IBL model is more robust than the SBL model. This exercise initiates a discussion of the SBL and IBL approaches to modeling choice and decision making in ACT-R and a reevaluation of how to compare and assess computational models.