Abstract: Ｒed light running at signalized intersections is a major safety concern in the United States. Statistics show that approximately 45 percent of crashes at intersections caused by red light running result in severe injuries and fatalities, while only approximately 30 percent of all other types of intersection crashes cause injuries or fatalities. Over the past decade, many US cities and counties have deployed red light running photo enforcement systems for signalized intersections within their jurisdictions to potentially reduce red light running related crashes. This study proposes an empirical Bayesian (EB) before-after analysis method that computes a weighed sum of crashes observed in the field and crashes predicted by safety performance functions (SPFs) to mitigate regression-to-mean biases for analyzing crash reduction effects of red light running enforcement. The analysis explicitly considers red light running related crash types, including head-on, rear-end, angle, turning, sideswipe in the same direction, and sideswipe in the opposite direction; and crash severity levels classified as fatal, injury, and property damage only (PDO). A computational study is conducted to examine the effectiveness of the Chicago program with red light running photo enforcement systems deployed for nearly two hundred signalized intersections. It is revealed that the use of red light running photo enforcement on the whole is positive, as demonstrated by reductions in all types of fatal crashes by 4-48 percent, and injury-related angle crashes by 1 percent. However, it slightly raises PDO-related angle crashes and moderately increases injury and PDO related rear-end crashes. The safety effectiveness of red light running photo enforcement is sensitive to intersection location.