Abstract: Soil erosion prediction technology has been proven to be a valuable tool for dealing with problems in the soil conservation and for advancing our understanding of soil erosion processes. Though substantial process has been made regarding the development and use of erosion models still significant improvements are needed to describe many fundamental processes and their spatial and temporal variations. In this paper some weakness associated with present-day models for soil erosion by water are discussed and some opinions on model development are proposed. It will be good for the next generation of erosion models. In the soil erosion models which processes to be represented and how to represent them are two principles. Different processes represented and diverse methods used lead to the diversity of soil erosion models with different properties. The water erosion models tend to be complex and physically based, but they do not achieve high accuracy correspondingly because of the error accumulation and data availability. At different scales different groups of processes are dominant and the effective focus of the model also changes with scales. It is important to keep represented and observed processes in the same spatio-temporal scale. The main reasons for the uncertainty of model performance are: complexity of natural processes and phenomena difficulties in obtaining appropriate input parameter values and built-in problems of the model. Models combined with GIS which is characterized of high accuracy and low data requirements are desired greatly. More attention should be paid to certain processes such as sidewall erosion, gravity erosion, and effect of crust and soil microbe and methods to obtain data more accurately. Improved ways of dealing with heterogeneity and scale are needed. The multi-scale and regional scale model will also be the study foci

Key words: soil erosion model, process, spatio-temporal scale, uncertainty