Traffic signal systems are usually retimed on the basis of sampled traffic counts. Yet limited traffic data may not represent typical conditions for optimizing traffic signal systems. When extensive traffic counts are not available, several approaches must be considered: How should signal timings be optimized if multiple counts exist? Should signal timings be based on maximal (or near maximal) traffic counts or should traffic volumes that are more frequently observed in the field be used? This study provides answers to those questions by investigating the performance of signal timing plans developed for various traffic count scenarios, at conditions of varying traffic demand. One key contribution of this study is accurate modeling of varying traffic conditions that were captured from field traffic counts during 155 weekdays in 2009. VISSIM's model of Park City, Utah, calibrated and validated in previous studies, was used to model field traffic conditions and to evaluate the quality of various signal timing plans. Results show that signal timing plans that are based on average traffic flows (mean, mode, and median) perform best (and are most robust) when exposed to day-to-day traffic flow variability. Results also show that although optimizing signal timings for higher traffic demand is suboptimal, this strategy is better than optimizing signals for lower traffic demand and should be used when sufficient traffic data are not available.