A parallel decentralized damage detection method is proposed for a large-scale structure with a short-term field test, with the structure divided into a number of small zones. Local responses in a zone are effectively used for the damage assessment of structural parameters in the same zone. The external excitations in each local test are identified in the wavelet domain and the unknown structural parameters are updated by using the Newton SOR method in the time domain. Both the external excitations and the structural parameters are updated iteratively with parallel computing. Both the external excitations and the structural parameters are updated iteratively with parallel computing. Numerical studies on a plane frame structure show that the method is effective for damage detection of the structure with a few sensors, even with 10% noise in the measurements. There is a gain of approximately 37% in the required computation time when parallel computing is used compared with results from computation in sequence.