We carried out experiments to detect impact locations on a composite plate using two types of composite plates, a composite flat plate with a constant thickness of 5 mm and a composite stiffened panel with stringers. Four multiplexed FBG sensors were attached to the bottom surface of the composite plates to acquire impact signals. The FBG sensor wavelength shift data were collected at a sampling frequency of 40 kHz using a high-speed FBG interrogator (SFI-710, Fiberpro Inc., Korea). The arrival times of the impact signals at each FBG sensor were obtained using a signal processing procedure. The arrival times were affected by noise level and signal-to-noise ratio. In order to overcome this weakness, signal processing techniques such as wavelet decomposition, normalization using each noise level and filtering with a moving average were adopted. To calculate the impact locations of the composite plate, a neural network algorithm was applied.