The Horton (1945) infiltration model of surface runoff and erosion is shown to be of much more limited geomorphic application than has been recognised hitherto. It is most applicable to clay badlands with low infiltration capacities and little weathered cover, and is one end-member of a wide spectrum of erosion models. The other end-member applies to slopes with high infiltration capacities and thick soil covers where throughflow dominates, and overland flow, with its attendant channel initiation, only occurs in a few restricted areas. Throughflow is capable of producing peaks in some river hydrographs, as is demonstrated both by controlled field experiments (Whipkey, 1965) and, by inference, by the “partial area” concept of runoff (Betson, 1964; Tennessee Valley Authority, 1963, 1964 and 1966). Experimental results of Hewlett (1961; Hewlett and Hibbert, 1963) are shown to be consistent with infiltration theory (Philip, 1957–8), indicating different response of throughflow to rainfall in the unsaturated upslope area and in the saturated zone at the base of slope near the channel. Areas of flow convergence, surface concavity and thinner or less permeable soil are shown to have higher moisture contents per unit volume, greater flows per unit cross-section and consequently, overland flow at lower rainfall intensities and durations than elsewhere. Rain falling on soil-covered drainage basin, in which throughflow dominates, will encounter a much more varied areal distribution of soil moisture than Horton postulated. Overland flow and surface erosion are most common, therefore, in locations marginal to the existing channel network and particularly in the hollows at stream heads. In these hollows the position of the channel tip is determined by a balance between fluvial cutting and basal fill along the valley axis by slope processes. Interpretation of stream-head hollows as areas of fill, periodically re-excavated when rare storms extend the drainage network, may also have a bearing on the origin of some small dry valleys. Drainage density is controlled by the extension of the drainage network which, in turn, is controlled by throughflow and overland flow near the foot of slopes and, particularly, in stream-head hollows. This implies that drainage density controls the behaviour of stream hydrographs, and not the reverse as postulated by Carlston (1963 and 1966). Non-linearity of stream hydrographs arises from variations in mean stream velocity with discharge (Pilgrim, 1966) and from non-linearity of the flow from soil-covered slopes. Because the rising limb of the hydrograph is fairly insensitive to the exact form of the slope outflow hydrograph, the linear unit hydrograph model usually provides a good approximation to peak flow. The recession limb of the hydrograph is much more sensitive to the slope outflow hydrograph and low stream velocities are more common after the flow crest. As a result, non-linearities become important and the unit hydrograph is an unreliable estimator of the form of the recession limb.