In this paper an attempt was undertaken to apply erosion-hydrological models, i.e. USLE (Universal Soil Loss Equation) and AGNPS (Agricultural Non-Point Source Pollution) for estimation and computer simulation of erosion risks and surface water pollution characteristics in the Ciemięga river catchment. Basing on the available literature data, about the studied catchment area as well as on the authors research, performed in the vegetation and hydrological periods of the years 1996-1997, the input data were collected to run the above models. They included data on: catchment physiography, land use, climatic conditions, soil features and human induced environment impacts. The study area is typical for loessial rolling terrain of the Lublin Upland. It is very susceptible to erosion processes which are additionally accelerated by agricultural activities. All the above factors have significant influence on surface water quality. The results on the content of micro- and macroelements in the Ciemięga river water evidenced that chemical loads were irregularly distributed along the river course, depending on the intensity and type of land use and human activity. The amounts of chemical elements measured in water in the study periods did not exceed the critical loads, thus they were not harmful to the environment and water quality, except the level of phosphates. Computer simulation performed for surface runoff and soil loss showed that USLE gives reasonably good data to estimate and forecast erosion rates and risks within the catchment in the whole hydrological year. Therefore, it can be stated that USLE model is sensitive enough for this general characteristics, despite the fact that many input data to compute this equation are of qualitative (not quantitative) character. The USLE simulation showed that the erosion intensity in the study area was most significantly influenced by the crop and land use types. AGNPS programme, which allows to calculate and estimate the rate and amount of surface runoff, outflow, soil and chemical elements losses at the moment of rain occurrence (single event), showed high sensitivity to rainfall parameters. It should be stated that simulation data were in accordance with both the literature and authors' experimental data. They allowed for complex estimation of erosion risks in the study area, i.e. lossial river catchment where agriculture activity predominates. Finally, it can be concluded that both USLE and AGNPS models were used for the first time to calculate and estimate erosion rates in loessial basins in Polish conditions, and because the simulation data appeared to be reliable and were in accordance with experimental data available so-far, they can be recommended for further considerations in other basins.