Following the wide and successful use of continuous shoulder rumble strips, many state departments of transportations (DOTs) installed centerline rumble strips (CLRS) on rural two-lane and undivided multilane highways in an effort to reduce cross-over-the-centerline (COCL) crashes. COCL crashes include head-on, sideswipe opposite direction, fixed object run-off-the-road left, and non-collision. The purpose of this research was to develop guidelines for using CLRS in Virginia based on a review of best practices and the analysis of Virginia COCL crash data from 2001 through 2003. The analysis procedures included data query and analyses of crash frequency, density, and rate. Areas and route locations with the highest COCL crashes and densities were identified as potential candidate sites for CLRS. As of 2003, 24 state DOTs and two Canadian provinces were using CLRS. They are generally installed on a case-by-case basis. CLRS design patterns vary greatly among states, but the most commonly used types are continuous grooves 12 to 16 inches in length, 6 to 7 inches in width, and 0.5 inch in depth spaced 12 or 24 inches apart. The optimal CLRS patterns remain unknown. Data analyses revealed that the distribution of COCL crashes in Virginia varied significantly with roadway system, road type, jurisdictional area, and road location. The statewide COCL crash densities were 0.13 and 0.71 crash per mile for secondary and primary roads, respectively. Fixed object run-off-the-road left was the predominant type of COCL crash followed by sideswipe opposite direction and head-on for undivided roads. The crash density of the primary system was 4.5 times higher than that of the secondary system. Guidelines were developed that outline the application of CLRS, design dimensions, installation and maintenance, and other issues. The authors recommend that the Virginia Department of Transportation's Traffic Engineering Division implement the guidelines as a division memorandum. Although a benefit-cost ratio for this recommendation will vary with each site, a sample estimated benefit-cost ratio was at least 7.6 per mile.