This paper compares a local and a nonlocal scheme for vertical diffusion in the atmospheric boundary layer (ABL) within the NCAR Community Climate Model (CCM2). The local scheme uses an eddy diffusivity determined independently at each vertical level based on local gradients of wind and virtual potential temperature. The nonlocal scheme determines an eddy-diffusivity profile based on diagnosed boundary-layer height and turbulent velocity scale, and incorporates nonlocal vertical transport effects for heat and moisture. The nonlocal scheme also couples directly to the parameterization of deep and shallow convection. The two schemes are compared with independent radiosonde observations for several locations. The focus is on temperature and humidity structure over the ocean, where surface temperatures are specified, as the boundary-layer scheme strongly interacts with the land-surface parameterization. Systematic differences are shown in global climate simulations using the two schemes. The nonlocal scheme transports moisture away from the surface more rapidly than the local scheme and deposits it at higher levels. The local scheme tends to saturate the lowest model levels unrealistically, leading to clouds too low in the atmosphere. The nonlocal scheme was chosen for CCM2 because it provides a more comprehensive representation of the physics of boundary-layer transport in dry convective conditions. The nonlocal scheme is more robust numerically, as it is less sensitive to stability oscillations. The local scheme was used in the previous version of CCM (CCM1). The purpose of this paper is to document the nonlocal ABL scheme and show its impact on the global climate simulation produced by CCM2. The impact is determined by comparing with the results of an updated local-K approach. The paper focuses on results for temperature, specific humidity, and low clouds. The nonlocal scheme produces better simulated specific humidity profiles over the ocean compared to the local scheme. The local scheme underestimates the total mixing of specific humidity, resulting in too moist atmospheric levels near the surface. This affects surface fluxes of latent and sensible heat and may result in unrealistically large amounts of low-level clouds. The temperature and specific humidity profiles are also influenced at heights above the ABL due to the interaction of the nonlocal scheme with other parts of the model. The nonlocal scheme is more effective in representing the physics of boundary-layer transport in dry convective conditions. It produces more realistic boundary-layer heights and better represents the vertical transport of heat and moisture. The nonlocal scheme also has a direct coupling to the convection parameterization, which improves the simulation of convection. The nonlocal scheme is more robust numerically and less sensitive to stability oscillations. The local scheme is less effective in representing the physics of boundary-layer transport in dry convective conditions and is more sensitive to stability oscillations. The nonlocal scheme produces more realistic boundary-layer heights and better represents the vertical transport of heat and moisture. The nonlocal scheme also has a direct coupling to the convection parameterization, which improves the simulation of convectionThis paper compares a local and a nonlocal scheme for vertical diffusion in the atmospheric boundary layer (ABL) within the NCAR Community Climate Model (CCM2). The local scheme uses an eddy diffusivity determined independently at each vertical level based on local gradients of wind and virtual potential temperature. The nonlocal scheme determines an eddy-diffusivity profile based on diagnosed boundary-layer height and turbulent velocity scale, and incorporates nonlocal vertical transport effects for heat and moisture. The nonlocal scheme also couples directly to the parameterization of deep and shallow convection. The two schemes are compared with independent radiosonde observations for several locations. The focus is on temperature and humidity structure over the ocean, where surface temperatures are specified, as the boundary-layer scheme strongly interacts with the land-surface parameterization. Systematic differences are shown in global climate simulations using the two schemes. The nonlocal scheme transports moisture away from the surface more rapidly than the local scheme and deposits it at higher levels. The local scheme tends to saturate the lowest model levels unrealistically, leading to clouds too low in the atmosphere. The nonlocal scheme was chosen for CCM2 because it provides a more comprehensive representation of the physics of boundary-layer transport in dry convective conditions. The nonlocal scheme is more robust numerically, as it is less sensitive to stability oscillations. The local scheme was used in the previous version of CCM (CCM1). The purpose of this paper is to document the nonlocal ABL scheme and show its impact on the global climate simulation produced by CCM2. The impact is determined by comparing with the results of an updated local-K approach. The paper focuses on results for temperature, specific humidity, and low clouds. The nonlocal scheme produces better simulated specific humidity profiles over the ocean compared to the local scheme. The local scheme underestimates the total mixing of specific humidity, resulting in too moist atmospheric levels near the surface. This affects surface fluxes of latent and sensible heat and may result in unrealistically large amounts of low-level clouds. The temperature and specific humidity profiles are also influenced at heights above the ABL due to the interaction of the nonlocal scheme with other parts of the model. The nonlocal scheme is more effective in representing the physics of boundary-layer transport in dry convective conditions. It produces more realistic boundary-layer heights and better represents the vertical transport of heat and moisture. The nonlocal scheme also has a direct coupling to the convection parameterization, which improves the simulation of convection. The nonlocal scheme is more robust numerically and less sensitive to stability oscillations. The local scheme is less effective in representing the physics of boundary-layer transport in dry convective conditions and is more sensitive to stability oscillations. The nonlocal scheme produces more realistic boundary-layer heights and better represents the vertical transport of heat and moisture. The nonlocal scheme also has a direct coupling to the convection parameterization, which improves the simulation of convection