At the present time, most ground-based radar estimations of ionospheric convection use observations from single-station facilities. This approach requires certain assumptions as to the spatial and/or temporal uniformity of the convection. In this paper we present a critical examination of the accuracy of these vector velocity determinations, using realistic modeled flow patterns that are time-stationary but not spatially uniform. We find that under certain circumstances the actual and inferred flow fields show considerable discrepancy, sometimes not even agreeing in the sense of flow direction! Specifically, we show that the natural curvature present in ionospheric convection on varying spatial scales can introduce significant error in the velocity estimate, particularly when the radius of curvature of the flow structure is less than or equal to the radar range to the scattering volume. The presence of flow curvature cannot be detected by radars which determine velocities from measurements in two viewing directions, and it might not be detected by radars using azimuth scanning techniques. Thus we argue that every effort should be made to measure the ionospheric convection by bidirectional or multidirectional observations of a common ionospheric volume and that a synthesis of coherent and incoherent radar observations from different sites is preferable to multidirectional single-station observations using either radar alone. These conclusions are applicable to any Doppler measurement technique and are equally valid for high-latitude wind patterns using Fabry-Perot interferometer techniques. ¿ American Geophysical Union 1991