Agent-based models can be used to simulate spatially-explicit animal behavioral processes and their interactions with human activities. This approach can be applied to predict the potential effects of such activities on animal behavior and individual condition that could lead, in turn, to alterations in vital rates and, ultimately, long-term population change. We developed an agent-based model to describe the effect of interactions with tourism on the behavior of bottlenose dolphins in Doubtful Sound (New Zealand). The model describes the temporal variation of the individuals’ hidden motivational states, the way in which these states interact to determine the activity of groups of dolphins, and the feedback influence of the group's activity on individual motivations and condition. Moreover, it realistically simulates the movement of dolphin groups in the fiord. The model also includes tour boat behavior, incorporating the way key geographical features attract these boats. In addition to tourism effects, we accounted for the spatial heterogeneity in both dolphin activities and shark predation risk. The final simulation platform generated a realistic representation of the social and behavioral dynamics of the dolphin and boat populations, as well as observed patterns of disturbance. We describe how this tool could be used to ensure effective management of the interactions between anthropogenic factors and bottlenose dolphins in Doubtful Sound, and how it could be adapted to evaluate the effects of human disturbance on other comparable populations. We then fitted the dolphin component of the model to data collected during visual studies of the Doubtful Sound dolphin population between 2000 and 2002 using a Bayesian multi-state modeling framework. However, when the parameter estimates from this fitting process were used in the agent-based model, biologically realistic representations of the population were not generated. Our results suggest that visual data from group follows alone are not sufficient to inform such agent-based models. Information on the spatial structure of the animals’ activities and an appropriate measure of individual condition are also required for successful model parameterization.