To investigate the acoustic emission (AE) precursor detection of landslide failures, a model test aiming at reproducing the typical shear surface deformation of different landslide modes was designed. The evolution characteristics of the AE signals were analyzed in terms of AE count, cumulative AE count, AE correlation diagrams, and corresponding time-frequency properties. The test results show that for the progressive deformation mode, the AE count experiences a low-level period, an active period and a rapid increase period, and the distribution of correlation diagram hits concentrates in a relatively small scale and then gradually scatters. There is low frequency signals firstly and then high frequency signals, and the energy proportion of the high-frequency signals shows an increasing tendency. For the sudden deformation mode, the magnitude of AE count increases sharply, leading to the cumulative AE count curve rises steeply, and correlation diagram hits distribution turns into relatively scattering rapidly. Furthermore, the high frequency signals and high energy proportion appear much earlier than that of the progressive deformation mode. For steady deformation mode, however, the acoustic emission activity is quite active in the initial stage, the cumulative AE count curve rises sharply and then maintains relatively flat trend, and correlation diagram hits distribution scatters firstly, then the signal hits distribution begins to concentrate in a relatively small scale. There are intensive high-frequency hits and high energy proportion earlier, and later they tend to decay in response to smaller magnitudes of movement. Comprehensive use of multiple features can help identify landslide deformation patterns more accurately under complex natural conditions, which may provide a promising reference for the field warning monitoring of the diverse landslide failures.