Guardrails are a key safety feature of modern roadways. Collisions with many roadside hazards, e.g. trees, poles and culverts, can be dangerous and guardrail prevents many crashes with such hazards. However, using guardrail safely and effectively is a challenging problem in itself. This research examined two aspects of the problem: 1) assessment of the injury risks posed by guardrail itself; 2) determination of appropriate guardrail length. When controlling for other factors, light truck / van / sport utility vehicles (LTVs) showed injury odds 3.9 times greater in end terminal crashes compared to guardrail face crashes, while cars showed no significant increase in injury odds. Additionally, the odds of injury in frontal end terminal crashes appeared to be between 3.9 and 5.0 times lower when the terminal design was compliant with the National Cooperative Highway Research Program (NCHRP) 350 crash testing protocol, compared to non-compliant designs. Rollover occurred in 10 % of all frontal guardrail crashes, and was initiated by the guardrail in roughly 46 % of instances. The evidence indicates that end terminal contact increases rollover odds by 6.9 times compared to guardrail face contact for LTVs, but not for cars. NCHRP 350 compliance of end terminals was not observed to have any significant effect on rollover propensity. In side-impact crashes with guardrail, end terminal crashes represented only about 25% of crashes but accounted for more than 70 % of the injuries sustained. End terminals compliant with NCHRP-350 may be about five times as safe as non-compliant designs, but the difference appears to be overshadowed by the high degree of risk involved in striking any narrow fixed object with the side of the vehicle. A somewhat larger sample appears necessary to make this finding significant at the 95 % confidence level. Only about 20 % of rollovers in non-tracking guardrail side crashes are initiated by contact with the rail; 80 % are initiated by some subsequent contact. Those rollovers which are rail-initiated appear to be about twice as likely to be initiated by a terminal as by the guardrail face. Cars showed odds of minor to severe injury 3.6 times greater than LTVs in end terminal crashes. End terminal designs compliant with NCHRP 350 were not observed to carry significantly different odds of minor to severe injury than non-compliant end terminals. The findings control for driver seat belt use, rollover occurrence, terminal orientation (leading/trailing), control-loss and the number of impact events. Rollover and non-use of seatbelts were observed to carry much larger increases in risk than end terminal type. For cars, electronic stability control (ESC) reduces odds of fatal crashes with roadside barriers by about 50 % For LTVs, ESC reduces barrier fatality odds by about 40 %. Based on the effectiveness levels observed in this research, it is estimated that ESC could prevent about 410 out of 1180 possible barrier-related fatalities per year by 2028, when 75 % of the fleet is estimated to be equipped with ESC. The study findings suggest that ESC significantly reduces road departures into roadside barriers, and/or that ESC changes departure conditions so that barrier crashes have less severe outcomes. This research has compared the current standard procedure for computing guardrail length of need (LON) with 'departure corridors' based on real-world road departure trajectories. Due to the current procedure's simplified treatment of road departure geometry, LON recommended by the current procedure becomes very conservative for hazards located closer to the roadside, and less conservative for hazards located further away. By contrast, the departure corridor technique developed in this research provides a known, precisely defined level of protection which remains the same for different hazard offsets. Departure corridors can be made for any desired level of protection, and the technique provides flexibility in how protection may be defined. Most importantly, the departure corridor technique is fundamentally more realistic than the current standard procedure and gives LON recommendations which provide protection levels that can be easily communicated to policy makers and other stakeholders. / Ph. D.