The delimitation is a frequent and danger failure in concrete-filled steel tube. The ultrasonic testing is an important method of NDT and it is usually used in detecting concrete-filled steel tube’s flaw .The ultrasonic signal include time parameter, frequency parameter, amplitude parameter and waveform parameter . At the present time the detection is mainly based on time parameter,not included the other parameters.The frequency parameter is sensitive to flaw in detecting.The artificial neural networks (ANN) is applied in order to enhance the detection's accuracy standard. Inputting the ultrasonic signal's time parameter and frequency parameter as characteristic number, through training and learning,we can indentify the failure's style. Though carrying out the combined diagnosing system into practical testing, the approving quantitative results is gained.