The most widely used engineering analysis and design techniques depend highly on having accurate linear time-invariant (LTI) models; therefore, developing LTI models of physical systems is very important. The purpose of this thesis is to present system identification techniques to produce linear time-invariant models for complex structures, in particularly flexible systems. The importance of improving upon the system identification techniques is emphasized, and a significant improvement in frequency domain least squares system identification is discussed. The coherence function is used to find frequency data points that are heavily corrupted by noise for the purpose of discarding them to improve the frequency domain system identification process. The Coherence Threshold technique is developed and examined with an illustrated example, followed with the application of the technique to data obtained from a real flexible system.