This article presents a performance optimization approach to incipient crack detection in a wind turbine rotor blade that underwent fatigue loading to failure. The objective of this article is to determine an optimal demarcation date, which is required to properly normalize active-sensing data collected and processed using disparate methods for the purpose of damage detection performance comparison. We propose that maximizing average damage detection performance with respect to a demarcation date would provide both an estimate of the true incipient damage onset date and the proper normalization enabling comparison of detection performance among the otherwise disparate data sets. This work focuses on the use of ultrasonic guided waves to detect incipient damage prior to the surfacing of a visible, catastrophic crack. The blade was instrumented with piezoelectric transducers, which were used in a pitch-catch mode over a range of excitation frequencies. With respect to specific excitation frequencies and transmission paths, higher excitation frequencies provided consistent detection results for paths along the rotor blade's carbon fiber spar cap, but performance fell off with increasing excitation frequency for paths not along the spar cap. Lower excitation frequencies provided consistent detection performance among all sensor paths. © 2013 The Author(s).