In order to test the strengths and limitations of cloud boundary retrievals from radiosonde profiles, 4 years of radar, lidar, and ceilometer data collected at the Atmospheric Radiation Measurements Southern Great Plains site from November 1996 through October 2000 are used to assess the retrievals of Wang and Rossow <1995> and Chernykh and Eskridge <1996>. The lidar and ceilometer data yield lowest-level cloud base heights that are, on average, within approximately 125 m of each other when both systems detect a cloud. These quantities are used to assess the accuracy of coincident cloud base heights obtained from radar and the two radiosonde-based methods applied to 200 m resolution profiles obtained at the same site. The lidar/ceilometer and radar cloud base heights agree by 0.156 ¿ 0.423 km for 85.27% of the observations, while the agreement between the lidar/ceilometer and radiosonde-derived heights is at best -0.044 ¿ 0.559 km for 74.60% of all cases. Agreement between radar- and radiosonde-derived cloud boundaries is better for cloud base height than for cloud top height, being at best 0.018 ¿ 0.641 km for 70.91% of the cloud base heights and 0.348 ¿ 0.729 km for 68.27% of the cloud top heights. The disagreements between radar- and radiosonde-derived boundaries are mainly caused by broken cloud situations when it is difficult to verify that drifting radiosondes and fixed active sensors are observing the same clouds. In the case of the radar the presence of clutter (e.g., vegetal particles or insects) can affect the measurements from the surface up to approximately 3--5 km, preventing comparisons with radiosonde-derived boundaries. Overall, Wang and Rossow <1995> tend to classify moist layers that are not clouds as clouds and both radiosonde techniques report high cloud top heights that are higher than the corresponding heights from radar.