In the network-based real-time kinematic (RTK) global positioning system approach, the rover positioning accuracy and reliability depends on the quality of the atmospheric corrections, which is largely a function of spatial and temporal variability of ionospheric and tropospheric parameters. The location of the rover receiver with respect to the reference network receivers is also a very important factor, especially for applications such as off-shore navigation, where favorable geometry cannot always be assured. The primary goal of this paper is to describe tests of the speed and reliability of the ambiguity resolution and the ultimate accuracy of kinematic positioning for two representative reference receiver geometries: (1) pentagonal reference receiver geometry, with network-rover separation up to 131 km, which represents a typical reference scenario where the rover is located inside the reference network; and (2) irregular geometry, simulating a shore-bound scenario where the reference network can support only extrapolation of the atmospheric corrections to an off-shore rover (outside the reference network), with network-rover separation up to 200 km. The latter scenario is of special interest here, as the objective is to investigate the maximum acceptable separation of the rover receiver from the shore-bound reference stations. The Ohio Continuously Operating Reference Stations (CORS) network is used to simulate both scenarios, and the MPGPS software developed at The Ohio State University Satellite Positioning and Inertial Navigation Laboratory is used to carry out the analyses over a 24-h period of varying ionospheric activity. As a result of this study, the error budget associated with both network geometries is obtained, and the limitations of the network approach as a function of the network-rover geometry can be ascertained.