To complete a permitted left turn successfully, a driver must first correctly assess the adequacy of a gap in the opposing traffic stream. A tendency to reject physically adequate gaps can lead to needless delay, while acceptance of an inadequate gap can lead to a conflict or a collision with an opposing vehicle. Traditional gap acceptance models use time to arrival as the main (and often only) determinant of whether a driver accepts a gap. Human factors research, however, suggests that drivers have difficulty in accurately perceiving time to arrival and that other variables, such as the distance to an oncoming vehicle, may be more salient. In the present study, video extraction methods were used to collect speed, separation distance, and time-to-arrival data for 74 drivers making permitted left turns at an urban intersection. Logit modeling of the drivers' gap choices indicated that distance to the opposing vehicle was the variable most reliably associated with gap selection, with speed and time to arrival making weaker contributions. All of the best models implied that the probability that a driver will accept a gap of a given duration increases as the speed of the oncoming vehicle increases.