The question of upstream separation of a current of dense rotating fluid, below a stagnant layer of lower density, that is critically controlled by a weir is addressed in laboratory experiments that follow from theoretical calculations. For a constant potential vorticity flow in a gradually varying channel, it is shown that a range of parameters lead to upstream separation. In all cases the flow reattaches before reaching the sill crest, so that a closed gyre is produced. In laboratory experiments separation and reattachment was observed in numerous cases where the theory predicts closed gyres. Although the shape and size of the gyre are in accordance with theory, the circulation within the gyre differs markedly from theoretical predictions. The mechanism that feeds energy into the gyre appears to be an important unsolved problem. The results indicate that upstream separation is a fact and that the dynamics of the gyre is significant, so the gyre dynamics may modify the volume flux of fluid over the weir. ¿ 1998 American Geophysical Union