This study conducted a before–after evaluation by means of the empirical Bayes methodology for four types of treatments at signalized intersections with data from Winston-Salem, North Carolina. The results indicated that changing to protected left-turn phasing from permissive or permissive–protected phasing could lead to a virtual elimination of left-turn crashes but other crashes, which were likely to be less severe, could increase. Conversion from nighttime flashback to regular phasing seemed effective in reducing nighttime crashes. Replacing 8-in. signal heads with 12-in. heads seemed effective in reducing right-angle crashes, but this measure could increase other, less-severe crashes. Adding another red-signal lens to an existing one or changing from permissive to permissive–protected left-turn phasing did not seem particularly effective in reducing crashes, but these results were not definitive because they were based on a limited number of sites. Further research using data from other jurisdictions is needed, so that more definitive conclusions can be made about the safety effectiveness of these treatments.