A central component of any discrete choice analysis is the selection of alternatives that determine a decision agent's choice set. Failure to properly specify choice sets will generate biased parameter estimates, inaccurate behavioral predictions, and erroneous estimates of policy relevant metrics (e.g., welfare effects of closed areas in fisheries). The development of more behaviorally realistic choice sets is integral to predicting agent behavior and informing public policy. In some contexts such as fisheries, discrete spatial choices are made repeatedly, and the decision-maker invests in the collection of fine-scale spatial information through time. For such data rich environments, we propose constructing choice sets by sampling from a fine-scale grid of location choices or from all observed tow starting points and compare this approach to a traditional conditional logit model with choices sets constructed of discrete fishing areas that aggregate many possible specific fishing locations. We present results from a Monte Carlo study that compares these modeling approaches in terms of parameter bias and prediction. We also compare results from an empirical application of the models to the Pacific Groundfish trawl fishery. We find considerable heterogeneity in the parameter estimates and support for our fine-scale choice set model in terms of both superior choice prediction and smaller parameter bias.