A method of fuel consumption extraction based on complex empirical mode decomposition is proposed to solve the fuel consumption extraction problems resulting from high quantization noise, and nonlinear and non-stationary characteristics of residual fuel volume recorded by a flight data recorder, as well as the mode mixing in empirical mode decomposition. First, a simulated signal with a similar morphology to the real signal is constructed with nonlinear support vector regression using the recorded signal’s key message. Second, the mode mixing can be reduced by using a simulant signal to synchronously guide the decomposition of the recorded signal in complex empirical mode decomposition. Finally, the fuel consumption is equivalent to the first-order derivative of real residual fuel volume estimated from the decomposition results. Simulation results show that this method has more advantages compared with other methods, and is adequate for extracting precious fuel consumption.