Abstract: An improved self-adaptive grey model based on wavelet denoising is studied to forecast the deformation of foundation trenches. As many factors may cause measured data to deviate from the real data during monitoring the deforming foundation trenches, a wavelet denoising method is first used to deal with the measured data of deformation. Then, the self-adaptive grey model is used to forecast the deformation with the initial value amended. An example based on the measured data of a foundation trench is given to show that deformation forecast results by the method combining the wavelet denoising and grey model is more precise than those by a conventional single grey model.