Abstract: Existing regional travel forecasting systems are not typically set up to forecast usage of bicycle infrastructure and are insensitive to bicyclists' route preferences in general. We collected revealed preference, GPS data on 162 bicyclists over the course of several days and coded the resulting trips to a highly detailed bicycle network model. We then use these data to estimate bicyclist route choice models. As part of this research, we developed a sophisticated choice set generation algorithm based on multiple permutations of labeled path attributes, which seems to out-perform comparable implementations of other route choice set generation algorithms. The model was formulated as a Path-Size Logit model to account for overlapping route alternatives. The estimation results show compelling intuitive elasticities for route choice attributes, including the effects of distance and delay; avoiding high-volumes of vehicular traffic, stops and turns, and elevation gain; and preferences for certain bike infrastructure types, particularly at bridge crossings and off-street paths. Estimation results also support segmentation by commute versus non-commute trip types, but are less clear when it comes to gender. The final model will be implemented as part of the regional travel forecasting system for Portland, Oregon, U.S.A.