Intersection collision warning systems can potentially reduce the number of collisions and associated losses. A critical design aspect of these systems is the selection of warning criteria, which represent a set of conditions and parameters under which the decision and the timing to issue warnings are determined. Proper warning criteria allow the generation of timely signals for drivers while minimizing false and nuisance alarms. We describe the development of a methodology to observe and analyze the selection of time gaps exhibited by driver behaviors in a real-world setting. The data collection procedures and analysis techniques are explained for left-turn across-path opposite direction scenarios, which constitute over a quarter of crossing path crashes at intersections. Exemplar data sets from an urban, signalized intersection are used to illustrate methods of deriving time-gap acceptance behaviors. The extracted information can serve as the basis for selecting gap acceptance thresholds in warning criteria, and the demonstrated methodology can be applied in the future development of intersection collision warning systems.