By making simple assumptions, an analytical theory is deduced for the mean velocity behind a two-dimensional obstacle (of heighth) placed on a rigid plane over which flows a turbulent boundary layer (of thickness δ). It is assumed thath[Gt ] δ, and that the wake can be divided into three regions. The velocity deficit −uis greatest in the two regions in which the change in shear stress is important, a wall region (W) close to the wall and a mixing region (M) spreading from the top of the obstacle. Above these is the external region (E) in which the velocity field is an inviscid perturbation on the incident boundary-layer velocity, which is taken to have a power-law profileU(y) =U∞(y−y1)n/δn, wheren[Gt ] 1. In (M), assuming that an eddy viscosity (=KhU(h)) can be defined for the perturbed flow in terms of the incident boundary-layer flow and that the velocity is self-preserving, it is found thatu(x,y) has the form$\frac{u}{U(h)} = \frac{ C }{Kh^2U^2(h)} \frac{f(n)}{x/h},\;\;\;\; {\rm where}\;\;\;\; \eta = (y/h)/[Kx/h]^{1/(n+2)}$, and the constant which defines the strength of the wake is$C = \int^\infty_0 y^U(y)(u-u_E)dy$, whereu=uE(x, y) asy→ 0 in region (E).In region (W),u(y) is proportional to Iny.By considering a large control surface enclosing the obstacle it is shown that the constant of the wake flow is not simply related to the drag of the obstacle, but is equal to the sum of the couple on the obstacle and an integral of the pressure field on the surface near the body.New wind-tunnel measurements of mean and turbulent velocities and Reynolds stresses in the wake behind a two-dimensional rectangular block on a roughened surface are presented. The turbulent boundary layer is artificially developed by well-established methods (Counihan 1969) in such a way that δ = 8h. These measurements are compared with the theory, with other wind-tunnel measurements and also with full-scale measurements of the wind behind windbreaks.It is found that the theory describes the distribution of mean velocity reasonably well, in particular the (x/h)−1decay law is well confirmed. The theory gives the correct self-preserving form for the distribution of Reynolds stress and the maximum increase of the mean-square turbulent velocity is found to decay downstream approximately as$ (\frac{x}{h})^{- \frac{3}{2}} $in accordance with the theory. The theory also suggests that the velocity deficit is affected by the roughness of the terrain (as measured by the roughness lengthy0) in proportion to In (h/y0), and there seems to be some experimental support for this hypothesis.