The complex full waveforms from laser altimetry, mixed with high noise, are usually reflected by the object with multiple height elevations. To accurately analyze the decomposition, vertical structure and characteristic parameters from these waveforms, a noise reduction method based on empirical mode decomposition (EMD) was investigated and tested with the full waveform of nonlinear and nonstationary signals obtained by GaoFen-7 space-borne laser altimetry. The reconstruction of an effective waveform signal was implemented through reverse superimposition of its intrinsic mode functions (IMFs) and the residual. And then different selection methods for these IMFs were compared, such as removed high frequency, threshold, wavelet and detrended fluctuation analysis (DFA). The results show that EMD-DFA1 and EMD-1 IMF have a higher noise reduction effect on these full waveforms, followed by EMD-Wavelet and EMD-Threshold. Finally, EMD-DFA1 was performed on the full waveforms with single peak, mixed peaks and multiple peaks. And the results show that EMD-DFA1 does well adaptability.