The precise detection of flaw echoes buried in backscattering noise caused by material microstructure is a problem of great importance in ultrasonic non-destructive testing (NDT). In this work, empirical mode decomposition (EMD) is proposed to deal with ultrasonic signal. A time-frequency filtering method based on EMD is designed to suppress noise and enhance flaw signals. Simulated results are presented,showing that the proposed method has an excellent performance even for a very low signal-to-noise ratio(SNR). The improvement in flaw detection was experimentally verified using stainless steel pipe sample with artificial flaws.