Back electromagnetic force (EMF)-based methods are commonly used for sensorless control of interior permanent magnet synchronous machines (IPMSMs) in medium and high speed range. The feature of high robustness to system structure and parameter uncertainties makes the sliding-mode observer (SMO) a promising candidate for rotor position estimation. In a practical drive system, because of physical limitations, e.g., sampling frequency and computational resource, it is challenging to obtain a perfect sinusoidal waveform for the back EMF by a SMO, especially in high speed range. As a result, the rotor position obtained from the estimated back EMF by using the traditional inverse tangent method will have nonnegligible oscillations. This paper proposes a novel algorithm, which uses the estimated rotor speed as a feedback signal with the conventional back EMF-based inverse tangent method to extract the rotor position. The proposed algorithm can effectively mitigate the oscillation and improve the dynamic performance of the SMO for rotor position estimation. The proposed algorithm is validated by simulations in MATLAB Simulink as well as experiments on a high-power IPMSM drive system.