In this paper we attempt to develop a simple algorithm to apply GPS observations to the short-intermediate-term earthquake prediction practices.By using the Load Unload Response Ratio（LURR） values at different stages,the earthquake potentiality in a seismogenic region is evaluated.The LURR time series fluctuats around 1.0 when the source media is in a steady state,and becomes quite high just before a large earthquake.To show the validity of the approach,the 2004 Parkfield earthquake in southern California is chosen as the example.We find that about half a year before the earthquake the LURR time series derived from the nearby GPS stations increased obviously;while from some far GPS stations,we can not observe similar anomaly.Therefore,the anomaly may be resulted from the regional crust stability change.The fact suggests that we can get new precursory anomaly to predict large earthquake by properly disposing GPS measurement data.