Pressure and tracer measurements in boreholes drilled to the bottom of Ice Stream B, West Antarctica, are used to obtain information about the basal water conduit system in which high water pressures are developed. These high pressures presumably make possible the rapid movement of the ice stream. Pressure in the system is indicated by the borehole water level once connection to the conduit system is made. On initial connection, here also called ''breakthrough'' to the basal water system, the water level drops in a few minutes to an initial depth in the range 96-117 m below the surface. These water levels are near but mostly somewhat deeper than the flotation level of about 100 m depth (water level at which basal water pressure and ice overburden pressure are equal), which is calculated from depth-density profiles and is measured in one borehole. The conduit system can be modelled as a continuous or somewhat discontinuous gap between ice and bed; the thickness of the gap delta has to be about 2 mm to account for the water-level drop on breakthrough, and about 4 mm to fit the results of a salt-tracer experiment indicating downstream transport at a speed of 7.5 mm s(-1). The above gap-conduit model is, however, ruled out by the way a pressure pulse injected into the basal water system at breakthrough propagates outward from the injection hole, and also by the large hole-to-hole variation in measured basal pressure, which if present in a gap-conduit system with delta = 2 or 4 mm would result in unacceptably large local water fluxes. An alternative model that avoids these objections, called the ''gap opening'' model, involves opening a gap as injection proceeds: starting with a thin film, the injection of water under pressure lifts the ice mass around the borehole, creating a gap 3 or 4 mm wide at the ice/bed interface. Evaluated quantitatively, the gap-opening model accounts for the volume of water that the basal water system accepts on breakthrough, which obviates the gap-conduit model. In order to transport basal meltwater from upstream it is then necessary for the complete hydraulic model to contain also a network of relatively large conduits, of which the most promising type is the ''canal'' conduit proposed theoretically by Walder and Fowler (1994): flat, low conduits incised into the till, similar to 0.1 m deep and perhaps similar to 1 m wide, with a flat ice roof. The basal water-pressure data suggest that the canals are spaced similar to 50-300 m apart, much closer than R-tunnels would be. The deepest observed water level, 117 m, is the most likely to reflect the actual water pressure in the canals, corresponding to a basal effective pressure of 1.6 bar. In this interpretation, the shallower water levels are affected by loss of hydraulic head in the narrow passageway(s) that connect along the bed from borehole to canal(s). Once a borehole has frozen up and any passageways connecting with canals have become closed, a pressure sensor in contact with the unfrozen till that underlies the ice will measure the pore pressure in the till, given enough time for pressure equilibration. This pressure varies considerably with time, over the equivalent water-level range from 100 to 113 m. Basal pressure sensors 500 m apart report uncorrelated variations, whereas sensors in boreholes 25 m apart report mostly (but not entirely) well-correlated variations, of unknown origin. In part of the record, remarkable anticorrelated variations are interspersed with positively correlated ones, and there are rare, abrupt excursions to extreme water levels as deep as 125 m and as shallow as 74 m. A diurnal pressure fluctuation, intermittently observed, may possibly be caused by the ocean tide in the Ross Sea. The lack of any observed variation in ice-stream motion, when large percentagewise variations in basal effective pressure were occurring according to our data, suggests that the observed pressure variations are sufficiently local, and so randomly variable from place to place, that they are averaged out in the process by which the basal motion of the ice stream is determined by an integration over a large area of the bed.
