This study examined 4,709 two-vehicle red-light-running (RLR) crashes at four-way intersections in California. Negative binomial models were developed to relate RLR crashes with geometric characteristics. Each one-lane increase in the width of the mainline street resulted in a 7 percent increase in RLR crashes for vehicles on the cross-street. However, the width of the cross-street did not affect the number of RLR crashes for vehicles on the mainline. RLR crashes for vehicles on the mainline street increased with higher ADT on both the mainline and the cross-street. In contrast, RLR crashes for vehicles on the cross-street increased with higher ADT on the cross-street but was unaffected by ADT on the mainline. The number of RLR crashes was 35 to 39 percent higher at intersections with fully-actuated signals than at those with pre-timed signals. These results can be used to target specific intersections for law enforcement measures.