Soil phosphorus tests offer a potentially powerful tool for land managers trying to predict the areas which will contribute diffuse losses of phosphorus (P) to surface water bodies through the overland flow vector - but do they work? We address this question at a range of scales, from patch (< 1 m2), through plot (several m2) to small watershed (several hectares). Our hypothesis is that as we increase the scale, and therefore the complexity of the system, soil P tests will predict P concentrations and losses associated with overland flow less well, and that this is partly due to a shift from dissolved P losses to P losses associated with eroded soil material. At the patch scale soil P tests were used to predict the P concentration and load from 24 European soils exposed to simulated rainfall under controlled conditions in the laboratory. Results showed that soil P tests were generally good predictors of reactive P <0.45 m, but did less well at predicting total P > 0.45 m. By combining the soil P test with measured sediment concentrations predictions of total P concentrations improved. Outdoor rainfall simulation experiments on bare soil plots (10 m2) revealed the overwhelming influence of particle bound P losses compared with P losses in the water phase. Soil P tests, which relate primarily to the dissolved P fractions in soil, were not able to predict total P losses, but were related to reactive P < 0.45 m losses. At the watershed scale soil P tests were able to predict reactive P < 0.45 m losses, but with considerable uncertainty. We conclude that soil P tests, in combination with sediment concentration provide a useful means of assessing the mobilization of P in overland flow, but should not be expected to provide watershed scale predictions of the movement of P into overland flow.