Water flowing in tubular channels inside a glacier produces frictional heat, which causes melting of the ice walls. However the channels also have a tendency to close under the overburden pressure. Using the equilibrium equation that at every cross-section as much ice is melted as flows in, differential equations are given for steady flow in horizontal, inclined and vertical channels at variable depth and for variable discharge, ice properties and channel roughness. It is shown that the pressure decreases with increasing discharge, which proves that water must flow in main arteries. The same argument is used to show that certain glacier lakes above long flat valley glaciers must form in times of low discharge and empty when the discharge is high, i.e. when the water head in the subglacial drainage system drops below the lake level. Under the conditions of the model an ice mass of uniform thickness does not float, i.e. there is no water layer at the bottom, when the bed is inclined in the down-hill direction, but it can float on a horizontal bed if the exponent n of the law for the ice creep is small. It is further shown that basal streams (bottom conduits) and lateral streams at the hydraulic grade line (gradient conduits) can coexist. Time-dependent flow, local topography, ice motion, and sediment load are not accounted for in the theory, although they may strongly influence the actual course of the water. Computations have been carried out for the Gornergletscher where the bed topography is known and where some data are available on subglacial water pressure.