In this paper we investigate the dynamic decision-making task of primary care physicians treating patients with type 2 diabetes to achieve a blood glucose goal. The focus of the study is on developing and testing an information processing theory that can explain why some physicians more often succeed and others more often fail to achieve desirable clinical goals. The developed theory is represented in the form of two types of computational models, one employing a feedback decision-making strategy and the other a feedforward strategy. The models were implemented in software and tested using data from a previously reported experiment where physicians treated simulated patients with type 2 diabetes. The physician data were scored for a defined set of treatment errors. Computational processes were systematically examined to identify and specify processes to perturb in order to generate the observed errors. Models were created for each physician by introducing perturbations in computational processes based on errors that each physician committed during the experiment. These models treated the same simulated patients that the physicians treated; results from each model treating the patients were compared with the represented physician’s results to test the sufficiency of the models to explain observed errors. Process perturbations which explained observed errors took two characteristic forms, both of which resulted in delayed treatment action: (1) elevated thresholds for triggering action and (2) overestimating delayed effects of medications. Physician models made predictions for types and timing of subjects’ treatment errors: physician models generated 79 % of the same types of treatment errors as committed by physicians. As demonstrated by this study, developing task specific information processing theories (expressed as computational models) are useful for investigating patterns of decision making that lead to errors of performance. Studies of this nature can support the design of decision support systems intended to reduce errors associated with dynamic tasks, such as treating a chronic disease.