A crash avoidance database structure that is based on driver judgments is proposed. The structure comprises four driving conflict states (low risk, conflict, near crash, and crash) that correspond with advisory warning, crash-imminent warning, and crash mitigation countermeasures. The feasibility of this database structure is investigated by answering two questions: (a) Can the driving states be reliably quantified? and (b) Can the quantified states be used to create a useful crash avoidance database? The feasibility discussion centers on a specific dynamic scenario that involved braking maneuvers by a following vehicle to avoid a rear-end crash with a stopped lead vehicle. The quantification of driver judgment data from a controlled test track study is discussed as a foundation to identify rough quantitative locations for the conflict and near-crash state transitions, and crash data from a driving simulator experiment are used to estimate the crash state boundary. A database of on-road, naturalistic driving data is compared with the controlled experiments to evaluate the results. The method is found to be feasible, and recommendations for further development are presented.