Operators of wind power facilities can mitigate wildlife mortality by slowing or stopping wind turbines (hereafter ‘curtail’) when birds are at an increased risk of collision. Some facility operators curtail when individual birds have flight characteristics (e.g. altitude, distance or relative bearing of a bird's flight path) that exceed some threshold value, but thresholds currently in use have not been empirically evaluated. Overly restrictive thresholds can cause turbine curtailment for birds that never enter rotor-swept zones, thereby resulting in excess power loss. We evaluated the probability that birds, specifically eagles, entered the rotor-swept zone (hereafter ‘entry probability’) in response to their flight characteristics. We used an automated monitoring system to classify individuals as eagles or non-eagles and record flight paths of purported eagles at a wind facility in Wyoming, USA. We used logistic regression with occupancy dynamics and a distance-dependent colonization process to model entry probability. As a result, this model allowed entry probability to decrease with horizontal distance to the nearest turbine. The probability of entry varied with distance to the nearest turbine and approached zero when that distance was more than 202 m. Entry probability peaked when eagles flew 89 m above ground, corresponding to hub heights of turbines (80 m), and decreased to near-zero at altitudes of 189 m or more. Entry probabilities were greatest when flight paths were near the rotor-swept zone and when eagles flew slowly toward the nearest turbine. Compass bearing of a flight path was not associated with entry probability. Our model accurately forecasted entry probability in Wyoming (area under the curve (AUC) = 0.96) and was transferable to another facility in California, USA (AUC = 0.97); therefore, our results may be applicable across a variety of settings. Curtailment criteria can be based on flight path characteristics to forecast entry into rotor-swept zones. The use of distance and altitude thresholds when making curtailment decisions is justified. However, this analysis suggests alteration of the time to collision threshold, with curtailment initiated at greater distances as the speed of the bird decreases. Our novel modelling method and our results can inform curtailment criteria in any situation where curtailment decisions are made in real-time.