In the field of permanent magnet synchronous motor (PMSM) control, the sliding mode observer (SMO)-based sensorless control is widely used; however, the actual control input of the current observation function is asymmetric. It can lead to different velocities of the estimated currents approaching to the actual currents and will make the current and back EMF fluctuations more severe, and result in more skewed angle and speed estimates, especially at a lower carrier ratio. In response to the above problems, this paper proposes a variable weighting coefficient of an EMF-based sliding mode observer (VWC-SMO). Unlike the traditional sliding mode observers, the weighted sliding mode switching variables and their bandpass-filtered values are used as the input of the current observer in the VWC-SMO. Thereby, the asymmetry of the control input in the current observation function can be well-suppressed, and almost the same approaching velocity on the two sides of the sliding surface can be obtained. Therefore, chattering near the sliding surface can also be suppressed. The method is verified on a motor controller experimental platform, and the comparative results shows that the VWC-SMO can reduce chattering of the observed currents and mitigate back EMFs fluctuations and improve the dynamic and steady-state performance.