In order to set up a rational land taxation system in China, a farmland classification and evaluation system claims precedance over all others. As one of the experimental study regions, Shihezi city is situated in the centre of the northwest arid desert area of China.Tis annual precipitation is little than 200 mm, and oasis agriculture is the main characteristics of the land use. The authors proceed from the specific land evaluation purpose and the regional features, and set up a farmland classsification and evaluation system which has the following distinguishing features:1. Formulate the calculation of the theoretical standard crop yield, thus providing scientific basis for differentiating of the farmland quality, accordingly we can avoid the errors caused by taking actual crop yield of every piece of farmland as the basis of the land quality evaluation. 2. Separately put 20 land quality influencing factors under three subsystems: the soil nutriment subsystem, water condition subsystem and the environment and technology subsyStem, by this we can accordingly introduce the model of soil nutriment equilibrium,the model of water circulation and the model of bioecosystem and social development into the process of farmland evaluation, thus greatly raising the quantitative extent of the evalu atio n.
3. Attach great importance to the analysis of water condition of each land unit. D "ring the form "lation of water condition index, we considered not only the normal factors which influence the balance of annual circulation of water resource, but also the factors, such as the water-yield reaction coefficients of each growth period of the standard crop, which influence the equilibrium between the water requirement and water supply in each different growth period in the life time of the crop.Finally, by comparing the integrated land quality index with the established standard, the farmland of Shihezi is classified into & different levels, the quality levels showed good correspondence relations with the differential land rent which were calculated by random sam piing investingation. The result has also showed the useful ness in the land use planning and managament.