Attempts to extend known two-dimensional results (Ursell, 1947) to the fully three-dimensional case can lead to unpredictable results. We show how the use of a variational approximation for a finite plane vertical barrier leads to apparently different results when different formulations are used. The reason for this is not so much that the method is wrong, but rather that several different limits are taken in the process, which are hard to control. We suggest an alternative matching scheme, based on Ayad and Leppington (1977), which holds for the case ka → ∞, l/a → ∞, kl → ∞, where / is the length of the barrier, a its depth and k the wavelength of the incident wave. The method is applied to a channel with impeding side walls, as a model of French's (1977) wave-energy device.