Abstract: The main characteristics of Lamb waves are multi-mode and dispersion, in any given excitation frequency, there are at least two Lamb wave modes. The dispersion characteristics of each mode make Lamb wave detection become very complex, so extraction for characteristic parameters of the signal and accurate signal interpretation is the key of Lamb wave detection. The signal of multi-mode Lamb wave detection was analyzed directly by HHT, and the conclusion was obtained that the approach of using multi-mode Lamb waves signal by EMD decomposition to extract the instantaneous frequency and instantaneous amplitude directly was not suitable. The noise and the mutual interference of Lamb wave modes was effectively eliminated by improved HHT, and the time-frequency clustering of the signal was good, and it was of benefit to characteristic parameters extraction and further analysis. Testing for different size and depth of the hole type artificial defects, the peak of instantaneous amplitude and the mean of instantaneous frequency were extracted respectively for linear regression analysis, it was discovered that the peak of instantaneous amplitude was a better characteristic parameter than the mean of instantaneous frequency for the detection of hole type defect. And the sensitivity of the change of the defect diameter could be calibrated by the regression line slope K of the peak of instantaneous amplitude, which could be used to predict the size of unknown defects.