Phase-locked loops (PLL) are increasingly used for position and speed estimation in speed sensorless control of permanent-magnet synchronous motors (PMSMs). In this paper, the saddle point characteristic phase-portrait is developed to explore the undesirable oscillation phenomena of PLL-based position and speed estimators for speed sensorless control. First, the large-signal model of the PLL-based position and speed estimator is established. Then, the saddle point phase-portrait method and Lyapunov-based argument are adopted to characterize its performance. It is revealed that the PLL-based position and speed estimator has infinite convergence regions. When the system state jumps from one convergence region to another, it will induce undesirable oscillations. Experimental tests are provided to verify the effectiveness of the proposed saddle point characteristic phase-portrait-based analysis.