In this study, data from continuous measurement of natural rainfall and runoff rates on nine 2 m(2) runoff plots were used to analyse the dynamics of infiltration and how it relates to the evolution of the soil surface state. All measurements were performed on two cropped sites in northern France on low slope loess soils, with low rainfall intensities. Crops represented were maize, wheat and sugar beet. Infiltrability was characterized at the scale of the plot and of the rainfall event by a single parameter, the constant infiltrability that yields an excess rainfall hyetograph with a volume equal to the measured runoff volume (phi index). Results indicated that: (1) even if the runoff plots were installed on two catchments where water erosion was important, the observed runoff volumes and intensities were very small. The runoff to rainfall ratio was typically less than 5% at the season to year scale; (2) the apparent infiltrability phi cannot be defined independently from rainfall intensity when runoff at the outlet of a plot comes from only a subset of the plot area. The relationship between rainfall intensity and apparent infiltrability can be used to estimate the relative runoff contributing area (approximate to 0.10-0.35 in ourcase); (3) the development and nature of surface crusting has a major influence on the apparent infiltrability: there is a progressive shift of phi toward low values when crusting develops, which is predominantly due to an increase in the relative contributing area. From a methodological point of view, failure to include the rainfall intensity dependence of infiltrability in runoff modelling could introduce large errors on runoff predictions. This relationship between rainfall intensity and apparent infiltrability should thus be analysed when possible, and soil surface state characterization should include more information about connection between the different parts of a plot and its outlet.