The effectiveness of soil loss and runoff predictions by soil erosion models can be determined by rainfall simulations or measurements under natural rainfall. This study was conducted to develop a set of statistics that would highlight differences in the performance of three soil erosion models: EUROSEM, WEPP and MEDRUSH. Rainfall (six rainfall intensities, two replicates), runoff, and soil loss data from experimental runoff plots at four locations in Hungary were used to find differences among models. The selected soil types are heterogeneous in their physical and chemical properties and are widely used for arable farming. Results showed that the three soil erosion models perform with varying effectiveness, depending on basic soil properties and rainfall intensities. The Hierarchical Cluster Analysis showed that the runoff and soil loss values, forecasted by the models have different correlations with changing soil water contents. Based on of the Mann-Whitney U test we concluded that EUROSEM model showed the best correlation with the measured data on the examined Cambisol. For the Regosol studied, using data measured under wet soil conditions, statistical analysis showed that only the MEDRUSH model estimated runoff well; runoff forecasts of the EUROSEM and WEPP models differed from measured data.