Although most of transient electromagnetic (TEM) apparatuses in China and abroad have taken various methods to suppress noise and improve SNR, noise may still remain in TEM data after stacking of all the periods. Stacking samples in time windows is an important technique for TEM data after stacking of all the periods. This paper proposes that Robust M Estimates methods can be applied to stacking samples in time windows. M Estimates are respectively adopted to process TEM data, and perform simulation and measurement in the field. The results and errors are investigated and compared.The experimental results show that M Estimates could effectively remove the Gaussian random noise and impulse noise, the M Estimates method is superior to the Arithmetic mean and geometric mean stack, and thus can improve the quality of data preprocessing.