Traffic crashes continue to be a major concern on the U.S roads. Rural highways represent the largest single class of highways in the United States, and account for approximately 80 percent of all paved highways. Two-lane highways compose of more than 85 per cent of all rural highways. Most of the existing studies on the deer-vehicle-crash (DVC) and run-off-road (ROR) crashes are based on crash reports only, and there are some conflicting findings on the major contributing factors to the crashes. In this study we investigate the most common types of crashes on two-lane rural highways. We find that the deer-vehicle-crashes and the run-off-road crashes represent the majority of the crashes in a sample of 1208 non-intersections crashes that occurred in 173.5 miles of roadways in the Ashtabula County, Ohio. The investigation procedure begins with identifying the high crashes sections, followed by field surveys, and statistical analysis. The statistical analysis showed that the percentage of the wooded area located less than 30 ft from the highway edge is the most important factor that can be used to distinguish the high DVC sections from the other sections. Similarly, the percentage of the non-forgiving shoulders is the most important factor to distinguish between the high ROR sections from the other sections. In addition, the correlation analysis showed weak connection between the high DVC sections and the high ROR sections. Field experiments have been conducted to collect additional data unavailable in the crash reports. The experimental results suggest that the depth of clear zone (by removing trees and bushes) in the wooded area should be no more than 40 ft on each side of the highway. To increase highway lights, the areas that need to be lit are between 27 to 38 feet from the side of the highway. The results also suggest that deer warning signs have little influence on the drivers to reduce speeds or switch from low to high beams. Future studies may include investigation of the effectiveness of the countermeasures for DVC. Additional work is also needed to study the behavioral factors of drivers in the ROR crashes once data are available.