The authors use large-eddy simulation (LES) to investigate entrainment and structure of the inversion layer of a clear convectively driven planetary boundary layer (PBL) over a range of bulk Richardson numbers, Ri. The LES code uses a nested grid technique to achieve fine resolution in all three directions in the inversion layer. Extensive flow visualization is used to examine the structure of the inversion layer and to illustrate the temporal and spatial interaction of a thermal plume and the overlying inversion. It is found that coherent structures in the convective PBL, that is, thermal plumes, are primary instigators of entrainment in the Ri range 13.6 [less than or equal to] Ri [less than or equal to] 43.8. At Ri = 13.6, strong horizontal and downward velocities are generated near the inversion layer because of the plume-interface interaction. This leads to folding of the interface and hence entrainment of warm inversion air at the plume's edge. At Ri = 34.5, the inversion's strong stability prevents folding of the interface but strong horizontal and downward motions near the plume's edge pull down pockets of warm air below the nominal inversion height. These pockets of warm air are then scoured off by turbulent motions and entrained into the PBL. The structure of the inversion interface from LES is in good visual agreement with lidar measurements in the PBL obtained during the Lidars in Flat Terrain field experiment. A quadrant analysis of the buoyancy flux shows that net entrainment flux (or average minimum buoyancy flux [Mathematical Expression Omitted]) is identified with quadrant [Mathematical Expression Omitted] motions, that is, warm air moving downward. Plumes generate both large negative quadrant [Mathematical Expression Omitted] and positive quadrant [Mathematical Expression Omitted] buoyancy fluxes that tend to cancel. The maximum vertical gradient in potential temperature at every (x, y) grid point is used to define a local PBL height, [z.sub.i](x, y). A statistical analysis of z[z.sub.i] shows that skewness of [z.sub.i] depends on the inversion strength. Spectra of [z.sub.i] exhibit a sensitivity to grid resolution. The normalized entrainment rate [w.sub.e]/[w.sub.*], where [w.sub.e] and [w.sub.*] are entrainment and convective velocities, varies as [ARi.sup.-1] with A [approximately equal to] 0.2 in the range 13.6 [less than or equal to] Ri [less than or equal to] 43.8 and is in good agreement with convection tank measurements. For a clear convective PBL, the authors found that the finite thickness of the inversion layer needs to be considered in an entrainment rate parameterization derived from a jump condition.