So il erosion prediction technology has been proven to be a va luable too l for dealing w ith prob lem s in the soil conservation and for advanc ing our understand ing o f soil erosion processes. Though substantial process has been made regarding the developmen t and use of erosion mode ls, still significant improvements are needed to describe many fundam enta l processes and their spatia l and temporal variations. In this paper some weakness associated w ith present-day models for so il erosion by w ater are d iscussed, and some op in ions onmodel developmen t are proposed. Itw illbe good for the next generation of erosionmodels. In the soil erosionmode ls, wh ich processes to be represented and how to represent them are tw o principles. D ifferen t processes represented and diversemethods used lead to the diversity of so il erosionmode lsw ith different properties. Thew ater erosionmode ls tend to be comp lex and physica lly based, but they do not ach ieve high accuracy correspond ing ly because o f the error accumulat ion and data ava ilability. A t different scales, different g roups o f processes are dom inan,t and the effect ive focus o f themode l also changesw ith sca les. Itis important to keep represented and observed processes in the sam e spat io-tempo ral scale. Them ain reasons for the uncerta inty o fmode l performance are: complex ity of natural processes and phenomena, diff icu lt ies in obta in ing appropriate input param eter va lues, and buil-t in prob lems o f the mode.l Mode ls combined w ith G IS, wh ich is characterized o f high accuracy and low data requ irem en,t are desired g reatly. More attention shou ld be pa id to certain processes such as sidew all erosion, g rav ity erosion, and e ffect of crust and so ilm icrobe,

and methods to ob tain data mo re accurate ly. Improved w ays o f dealing w ith heterogene ity and sca le are needed. Themu lt-i sca le and reg ional sca lemode lw ill also be the study foci