Abstract Presented is the approach, implementation, and evaluation of two-way nesting in a split-implicit ocean model, the Navy Coastal Ocean Model (NCOM). Emphasis is on the strategies applied to feed back fields from the fine-mesh nest (child grid) to the coarse mesh (parent grid). On an appropriate separation of dynamic and feedback interfaces, attention is especially needed for the feedback interface of surface elevation. One particular issue addressed is the inconsistency between the 3D baroclinic velocities and 2D barotropic transports in the feedback. The discrepancy is inherently associated with bathymetry, depth integration, and the need to average over spatial grid points. A simple remedy is proposed and proven to be effective and necessary in realistic coastal applications. In addition to the full two-way nesting, a simplified two-way nesting approach is provided in which only the temperature and salinity are fed back from the nest, and the velocity fields are assumed to self-adjust according to the geostrophic balance. The performance of both approaches is evaluated using the idealized benchmark, propagation of a baroclinic vortex, and an application to the Mississippi River outflow in the northeast Gulf of Mexico, including a comparison with available observations. Discussions are also made on the computational efficiency of the two-way nesting and its sensitivity to the open boundary conditions in regard to noise suppression. Significance Statement The two-way nesting approach reported here can be adapted to other structured-grid ocean models, in particular those using the split-implicit technique. The treatment of the feedback interface for surface elevation is especially important for suppressing the noise production and improving the feedback consistency. An effective procedure is given to amend the inconsistency in the velocity field feedback that is inherently due to bathymetry.