A new turbulent mixing scheme, described in Part I of this paper, is tested in the climate and mesoscale configurations of the U.K. Met. Office’s Unified Model (UM). In climate configuration, the scheme is implemented along with increased vertical resolution below 700 hPa (the same as that in the mesoscale model), in order to allow the different boundary layer types and processes to be identified and treated properly. In both configurations, the new boundary layer (PBL-N) mixing scheme produces some improvement over the current boundary layer (PBL-C) scheme. The PBL-N scheme is able to diagnose different boundary layer types that appear to be consistent with the observed conditions, and the boundary layer structure is improved in comparison with observations. In the climate model, the boundary layer and cloud structure in the semipermanent stratocumulus regions of the eastern subtropical oceans are noticeably improved with the PBL-N scheme. The deepening and decoupling of the boundary layer toward the trade cumulus regime is also simulated more realistically. However, the cloud amounts in the stratocumulus regions, which were underestimated with the PBL-C scheme, are reduced further when the PBL-N scheme is included. Tests of the PBL-N scheme in the UM single-column model and in a development version of the UM, where the dynamics, time stepping, and vertical grid are different from the standard version, both show that realistic stratocumulus cloud amounts can be achieved. Thus, it is thought that the performance of the PBL-N scheme in the standard UM may be being limited by other aspects of that model. In the mesoscale model, improvements in the simulation of a convective case are achieved with the PBL-N scheme through reductions in layer cloud amount, while the simulation of a stratocumulus case is improved through better representation of the cloud and boundary layer structure. Other mesoscale model case studies show that there is a consistent improvement in fog probabilities and forecasts of cloud-base height. The rootmean-square errors in screen-level temperature are also reduced slightly. The weak daytime bias in wind strength is improved greatly through a systematic increase in the 10-m wind speed in unstable conditions. As a result of these trials, the scheme has been implemented operationally in the mesoscale model.