The load/unload response ratio (LURR) aims to identify a physical parameter, which reflects the damaging process in the seismogenic zone, and to use this parameter to predict an earthquake. In the seismically quiet period, the LURR value fluctuates around 1. Nevertheless, when the rock is close to failure or just before a strong earthquake, the value rises significantly, and it is much larger than 1. The LURR probably decreases sharply and quickly before the main shock. Therefore, the anomalous increase in the LURR time series covering a time interval of months to years may be efficiently used in earthquake prediction. In the present study, the spatio-temporal evolution of LURR is analyzed before moderate earthquakes from 1990 to 1999. Based on the LURR theory, the seismogenic integral is deduced, which is applied to predict the earthquakes during the studies interval. The results show that high Y values are observed with an elliptical or circular spatial distribution around the epicenter before the main earthquake. In addition, earthquakes usually occurred during periods of abnormally high fluctuations. In retrospective historical earthquake studies, the future earthquakes are usually located in the abnormal zones or at the edge abnormal areas. However, there is a significant difference in terms of the starting time, and the amplitude and area of abnormity are different because of the different magnitude and seismogenic zone. An obvious anomaly, with a probability of approximately 78.8%, is observed within 200 km from the epicenter before moderately strong events. In addition, compared with the theoretical time, the statistical time from the peak to the earthquake is shorter, which may indicate that the tectonic activities in Xinjiang are intensive, the pregnant period is short, and the rate of the shear strain is large.