Current terrain classification methods using waveforms of a laser altimeter are mainly based on machine learning, which is an empirical model. Starting from the theoretical model of the laser waveform, by deriving the analytic model of laser waveforms of sea surface and sea-ice surface, the sum of the amplitude differences between the sea waveform and sea-ice waveform was calculated according to one by one sampling point based on the temporal-distance weighting; and the sea or sea-ice waveform was classified by the calculated difference value. The terrain types corresponding to GLAS laser footprints in the north sea-ice area of Greenland were obtained by analyzing the point clouds captured by an airborne LiDAR, and in this area, the GLAS measurement waveforms were used to validate based on the classification method proposed in the paper. The results indicate that after eliminating the effect of saturated waveforms, the overall accuracy OA is better than 95% and the Kappa coefficient is approximately 0.89, which achieves a high classification accuracy. The new method extends the terrain type classification from taking the machine learning as basis to taking the semi-theory analytic model as basis for a laser altimeter, which can be an important reference to the terrain type classification based on laser waveform successively.