AbstractRill bank collapse is an important component in the adjustment of channel morphology to changes in discharge and sediment flux. Sediment inputs from bank collapse cause abrupt changes in flow resistance, flow patterns and downstream sediment concentrations. Generally, bank retreat involves gradual lateral erosion, caused by flow shear stress, and sudden bank collapse, triggered by complex interactions between channel flow and bank and soil water conditions. Collapse occurs when bank height exceeds the critical height where gravitational forces overcome soil shear strength. An experimental study examined conditions for collapse in eroding rill channels. Experiments with and without a deep water table were carried out on a meandering rill channel in a loamy sand and sandy loam in a laboratory flume under simulated rainfall and controlled runon. Different discharges were used to initiate knickpoint and rill incision. Soil water dynamics were monitored using microstandpipes, tensiometers and time domain reflectometer probes (TDR probes). Bank collapse occurred with newly developed or rising pre‐existing water tables near rill banks, associated with knickpoint migration. Knickpoint scour increased effective bank height, caused positive pore water pressure in the bank toe and reduced negative pore pressures in the unsaturated zone to near zero. Matric tension in unsaturated parts of the bank and a surface seal on the ‘interrill’ zone behind the bank enhanced stability, while increased effective bank height and positive pore water pressure at the bank toe caused instability. With soil water contents >35 per cent (sandy loam) and >23 per cent (loamy sand), critical bank heights were 0·11–0·12 m and 0·06–0·07 m, respectively. Bank toe undercutting at the outside of the rill bends also triggered instability. Bank displacement was quite different on the two soils. On the loamy sand, the failed block slid to the channel bed, revealing only the upper half of the failure plane, while on the sandy loam the failed block toppled forwards, exposing the failure plane for the complete bank height. This study has shown that it is possible to predict location, frequency and magnitude of the rill bank collapse, providing a basis for incorporation into predictive models for hillslope soil loss or rill network development. Copyright © 2006 John Wiley & Sons, Ltd.