[Background] Deep percolation (DP ) is termed as the movement of water from soil surface to sufficient depths, usually below the root zone. It generally occurs when infiltrated water exceeds the storage capacity of the soil and plays an vital role in hydrologic cycle. ccurate evaluation of the deep percolation is crucial to analyze the processes of the hydrologic cycle. Furthermore, deep percolation is pivotal for the management and rational evelopment of groundwater resources, especially in arid and semi-arid regions where water resources is deficient. [Methods] This paper reviews several approaches to assess deep percolation in arid and semi-arid regions, i. e. , empirical, physical, tracer, and numerical modelling. [Results] The principle, applicability, merits and drawbacks of the above- mentioned four approaches are commented. Due to the empirical coefficient requires calibration, the empirical pproach is limited while it is applied in the other regions. Physical approach includes lysimeter method, soil water flux method, water balance method, Darcy method, and underground water-tablefluctuation method. The tracer approach is used to estimate deep percolation by the identification of peak value, profile shape, and the amount of tracers, but this approach cannot directly measure the deep percolation; moreover, the spatial variation in tracers is not considered in it. Theoretically, the approach of numerical modelling can be used to estimate and predict the deep percolation under any ircumstance. Nevertheless, it is difficult to obtain the parameters that is necessary for the numerical modelling. [Conclusions] Considering the advantages and isadvantages of each method, the integration of existing evaluation methods and mutual verification of them can improve the precision of the simulation. Due to the spatial and temporal variability of DP , the integration of existing methods and GIS may evaluate the spatial heterogeneity of DP at large scale. Simultaniously, long-term series of field observation may not only acquire the dynamic information of DP , but also provide the data support for the parameters of each approach.