Abstract: The method of empirical mode decomposition (EMD) combined with cross-correlation difference-Duffing oscillator (CD-EMD) was proposed for weak signal detection of initial leakage in boiler pressure piping. By adding the pneumatic sound signal of 0.6 MPa to the furnace background noise signal, weak leakage detection was numerically simulated, and the detection results were compared between EMD and wavelet transform under fixed signal to noise ratios. Detection results were also compared between EMD combined with Duffing oscillator (D-EMD) and CD-EMD by reducing the signal to noise ratios. Results show that with EMD method, the signal to be detected can be identified from the furnace background noise at -5 dB signal to noise ratio, and the effect is obviously better than wavelet transform; at -24 dB signal to noise ratio, the signal cannot be identified normally by D-EMD method due to the serious interference of initial phase; CD-EMD method can effectively overcome the misjudgment caused by initial phase, which therefore can improve the real-time property of judgement by setting the empirical threshold.