This study aimed to measure the preventive effect of seat belt on traumatic brain injury (TBI) and to compare the effect according to the crash severities and collision directions. Korea In-Depth Accident Study (KIDAS) has collected vehicle and demographic data on injured occupants involved in motor vehicle collisions (MVCs) who visited three emergency medical centers for calendar years 2011-2016. Primary and secondary end points were TBI (abbreviated injury score 2+) and in-hospital mortality. Crush extent (CE) was classified into 1-2, 3-4, 5-6, and 7-9 according to the crash severity. We calculated adjusted odds ratios (ORs) of seat belts and CE for study outcomes and developed an interaction model in each collision direction using multivariate logistic regression analysis. Of the 2,245 occupants who were injured in MVCs, 295 (13.1%) occupants sustained TBI. In univariate analysis, old age, unbelted status, lateral collision, and higher CE were factors associated with TBI in MVCs. Occupants with belted status was less likely to have TBI and in-hospital mortality compared with those with unbelted status [AORs (95% CI) 0.48 (0.37-0.62) and 0.49 (0.30-0.81), respectively]. In interaction analysis, preventive effects of seat belts on TBI from MVCs were retained within CE 5-6 in frontal MVCs and within CE 1-2 in near side lateral MVCs, and those of seat belts on in-hospital mortality were reserved within CE 3-4 in frontal and rollover MVCs. The preventive effects of seat belts on TBI and in-hospital mortality are preserved within a limited crash severity in each collision direction.