High-occupancy vehicle (HOV) lanes have been deployed as a tool for traffic management in urban freeway systems to improve reliability and mobility of trips. As they are planned to traverse crowded urban areas, it is often difficult to acquire sufficient right-of-way for retrofitting HOV lanes to existing freeway systems with recommended cross-sectional design. The present study proposes a methodology to determine the optimal set of cross-sectional design for safety performance by evaluating individual impact of each design element on safety as well as tradeoffs between them. Detailed collision data of concurrent-flow buffer-separated HOV lanes along with their geometric features and traffic flow data were analyzed to estimate collision predictive models for HOV and the adjacent general purpose lanes by injury types. These models were used to determine the set of cross-sectional design elements that minimizes the expected collision occurrences. As a case study, a real freeway corridor where converting continuous to buffer-separated types was underway was selected to demonstrate the applicability of the proposed method. This case study shows that the method can assist to determine cross-section design of HOV facilities for safety based on currently available geometric space.