Accurate Global Navigation Satellite System (GNSS) satellite yaw modeling is critical for precise GNSS orbit determination and GNSS-based high-precision applications. The GNSS satellite’s nominal model is correct and works well most of the time. However, errors can still arise when the satellite is in eclipse and the beta angle approaches zero degree. Due to the binary nature of the yaw direction in the model during yaw maneuver in this period, an otherwise harmless small error in evaluating the yaw angle can lead to a wrong turn direction in the model and result in one wavelength of phase measurement error. Reverse Kinematic Precise Point Positioning (RPP) uses a network of ground receivers to dynamically estimate the XY offset of a GNSS satellite’s antenna phase- center from the satellite’s center of mass. Provided that the phase center is sufficiently offset from the satellite’s yaw axis to be observed, the actual yaw angle may then be recovered. This is in contrast to yaw angles modeled and/or estimated during precise orbit determination (POD). The RPP technique is routinely used at JPL to monitor the actual yaw attitude of GPS satellites (except Block IIR – small antenna offset) and helps validate our yaw attitude models. Due to the decommissioning of the large majority of Block II/IIA satellites, our operational process does not allow us to address the II/IIA yaw models. To overcome this limitation and reevaluate the II/IIA/IIF yaw models, we have reanalyzed 18 years of satellite yaw maneuvers for GPS block II, IIA and IIF satellites using the most recent GPS data reprocessing campaign conducted at JPL. Based on these long time-series of RPP and POD derived yaw angles, we document the discrepancies observed during yaw maneuvers for Block II, IIA and IIF satellites, in particular in the vicinity of the zero beta angle. We discuss approaches that could be implemented to mitigate errors in the prediction of the correct turn direction.