In order to study the relation between heavy rainfall, shallow pore pressures, and slope stability in hillslopes susceptible to debris flows, we have been observing debris flows and measuring rainfall and hillslope pore pressures in a 10-km2 study area in the central Santa Cruz Mountains near La Honda, California. A simple numerical model, based on the physical analogy of a leaky barrel, can simulate significant features of the interaction between rainfall and shallow-hillslope pore pressures. In the model, the barrel is filled at a rate equal to the rainfall intensity and drained at a rate equal to the product of the water level retained in the barrel, Z, times the drainage coefficient, kd. If the retained rainfall exceeds a critical water level, Zc, the slope becomes unstable. Thus, the threshold for the intensity and duration of storm rainfall required to initiate debris flows is determined by the kd and Zc values of the most susceptible slopes in the study area. Although comparisons of leaky-barrel-model simulations with piezometer measurements indicate reasonable agreement, the kd values of individual piezometers in the study area vary widely because of spatial variations in soil properties and slope geometry. The spatial distribution of rainfall, in contrast, is fairly uniform over a scale of a few square kilometers. Thus, rather than attempt to model an actual slope in any detail, the leaky-barrel model may be used to compare the relative hydrologic effects of storms by plotting the maximum response level, Zmax, against the drainage coefficient, kd. Values of Zmax versus kd can then be compared for storms that triggered debris flows versus storms that failed to trigger debris flows. Using this technique, we can backcalculate kd and Zc values for the most susceptible slopes in the study area. These threshold values are kd = 0.85/hr and Zc = 8.5 mm. This leaky-barrel-model threshold is consistent with, but slightly higher than, an earlier, purely empirical, threshold. We can also relate the number of debris flows triggered by a storm to the time and amount by which the leaky-barrel-model response exceeded the threshold during the storm.