Abstract: To solve the problem of corrosion acoustic emission (AE) source feature extraction and recognition, the AE detection validation and the amplitude-frequency characteristics were derived in theory. The amplitude of AE signal generated by bubble burst is proportional to its radius square and the liquid level, while the AE signal frequency is inversely proportional to the bubble radius. The AE signal amplitude of the steel and the steel corrosion product cracking is proportional to the local stress, and the AE signal frequency is proportional to the crack propagation velocity and inversely proportional to the crack propagation distance. Three Q235 specimens were separately immersed in the solution of 10%FeCla₃.6H₂0, 10%FeC1₃.6H₂0 with 0.01 mol/L HCL and 5%CuSO₄-5H₂0. The AE systems with high frequency and low frequency were used to detect the whole corrosion process AE signals. The AE signals of Q235 steel and the steel corrosion products cracking were detected as the verification experiment. The AE signals from different sources could be distinguished by AE hits count and the power spectrum. It is coincident with theoretical analysis. These conclusions have significant guidance for the corrosion detection and evaluation by on-line acoustic emission detecting.