In this paper, an autoregressive spectral analysis method is proposed for on-line structural damage detection. Firstly, the monitoring strain data are modeled as ARMA models. The power spectral density (PSD) value is directly determining by center-autocorrelation function and statistical autocorrelation function. Structural damage detection is conducted by comparing the values of PSD of the inspected structure. Two experimental applications to the detection of damage in industrial piping structure demonstrate the potential and effectiveness of the new damage detection technique.