ABSTRACTFlow of ice over a pair of streamlined bumps was studied in a laboratory apparatus. Experimental conditions were similar to those under a glacier about 100 m thick with a shear stress of 0.1 MPa. Temperatures in the ice and bed were monitored with thermistors. Cumulative deformation was determined with the use of threads frozen into the ice. The threads were initially vertical, but were deformed during the experiment.About 75 % of the accommodation of the bumps was by plastic deformation, and 25 % was by pressure melting. Much of the heat for melting came from the boundaries of the apparatus so the melt water formed did not refreeze in the lee of the bumps.The cumulative deformation in the ice is compared with a theoretical prediction of Lliboutry and Ritz (1978) for flow over hemispherical bumps. The principal difference between theory and experiment, aside from that in bump geometry, is that the theory assumes a frictionless boundary between the bed and the ice, whereas in the experiment small-scale roughness elements result in drag. Because the theory ignores this drag, it overestimates the velocity immediately above the bump. The experimental deformation profiles are consistent with those predicted for a non-linear rheology, but not with those predicted for a linear rheology.