Rainfall redistribution by plant canopy, notably the water flow down the plant stem (stemflow), modifies the incident rainfall rate at the soil surface and may affect runoff generation. To test this hypothesis, we observed and measured runoff at the plant scale with banana cultivated on tropical Andosol. Observation of runoff by video and matrix potential monitoring showed that, during a runoff event, the matrix potential increased mainly downstream from the pseudostem in line with the slope, delimiting a saturated zone of runoff propagation that appeared on video monitoring. The results indicate that rainfall redistribution by plant canopy, i.e. stemflow and dripping areas, enhances runoff even on soil with a high infiltration rate (mean hydraulic conductivity at saturation Ks of 67 mm h-1). Data analysis of 40 runoff events showed that events were composed of at least two runoff phases characterized by an abrupt increase in runoff coefficient (RC) from 0.16 to 0.65 between the first and the second phase. The change in RC was related to rainfall rate. Also, between the first and the second runoff phase, the apparent infiltration rate at the plot scale decreased from 30 to 10 mm h-1. This was related to an increase in runoff contributing areas (RCA), from an estimated 18% to 93% of the plot surface. However, data analysis and model simulations showed that the increase in mean rainfall rate in RCA due to stemflow was not sufficient to account for large runoff volumes. Hence, one must also take into account the spatial variation of hydraulic conductivity at saturation with low values relative to RCA (estimation for the second runoff phase was 7.6 mm h-1). Moreover, simulation results implied Ks decreases with time. Finally, rainfall redistribution may have an impact at a larger scale. In banana plantations, the hydraulic connectivity of runoff areas can enhance the stemflow effect up to the plot scale. From this point of view, the two-compartment scheme we chose for m