A new method based on sample entropy algorithm that can perform real-time detection of sub-synchronous oscillation is proposed. The rotor torque signal is chose to be analyzed. The sample entropy of rotor torque during continuous period is calculated. Sub-synchronous oscillation detection is achieved by analyzing the variation of sample entropy. The method needs less data and has good anti-interference ability. The effectiveness of the proposed method is proven by the result of the IEEE first benchmark system model simulation.