In this paper a denoising strategy based on the EEMD (Ensemble Empirical Mode Decomposition) is used to reduce the background noise in non-stationary signals, which represent the forces measured in scaled model testing of the emergency water landing of aircraft, generally referred to as ditching. Ditching tests are performed at a constant horizontal speed of 12 m/s and a vertical velocity at the beginning of the impact of 0.45 m/s. The measured data are affected by a large amplitude broadband noise, which has both mechanical and electronic origin. Noise sources cannot be easily avoided or removed, since they are associated with the vibrations of the structure of the towing carriage and to the interaction between the measurement chain and the electromagnetic fields. The EEMD noise reduction method is based on the decomposition of the signal into modes and on its partial reconstruction using the residue, the signal-dominant modes and some further modes treated with a thresholding technique, which helps to retain some of the sharp features of the signal. The strategy is developed and tested first on a synthetic signal with a superimposed and known background noise. The method is then verified on the measurement of the inertial force acting on the fuselage when it is moving in air, as in this case the added mass is negligible and the denoised force should equal the product of the mass by the acceleration, both of them being known. Finally, the procedure is applied to denoise the forces measured during the actual ditching experiments. The results are superior to those obtained by other classical filtering methods, such as a moving average filter and a low-pass FIR filter, particularly due to the enhanced capabilities of the EEMD-denoising strategy here developed to preserve the sharp features of the signals and to reduce the residual low-frequency oscillations of spurious origin.