A series of laboratory experiments are described in which the following major features of the flow field were observed. Well above the outlet the flow was essentially one of uniform vertical velocity, which is such that the free surface falls at a rate determined by the mass flux through the outlet, the isopyenics remaining horizontal. The small vertical velocity is converted to a considerably larger horizontal velocity in an essentially horizontal layer near the level of the outlet slot. The width of this withdrawal layer was almost constant over a large portion of the tank (except for the region near the outlet), and the velocity field within it was found to be steady after an initial period of establishment. Also the horizontal velocity at a given level in the withdrawal layer was found, to a good approximation, to vary linearly with the distance along the tank, and the velocity distribution, at a given station, was determined principally by the viscous stress, once a representative length had been established. For flows initiated in a uniform tank by suddenly opening a valve in the outlet line, the width of the withdrawal layer seemed to be uniquely determined on a scale, dependent on the flux, that appears to derive from terms that are negligible once the steady flow has been established. By placing suitable obstructions in the tank it was possible to obtain similar flows, but with various widths. We were also able to change the structure of the withdrawal layer by controlling the way the mass flux was brought to its final value, thereby establishing that the width of the withdrawal layer was dependent on its history.