Continuous Rapid Deformation (CRD) may be an important precursor before some strong earthquakes with magnitude greater than %M%s5.0. This paper shows some characteristics of CRD, and proposes a method for identifying precursory CRD that is closely related to the process of earthquake preparation. From the deformation data in Yunnan Sichan provinces, we obtained the tilt and strain thresholds of 4.5ms/d and 5 ×10 -7 /d, respectively. Compared the tilt CRD in this region before 9 earthquakes (short term fore shocks and after shocks not included) with magnitude greater than 5.8 in the period from January of 1990 to May of 2001, we found that 8 earthquakes occurred within 5 months after the CRD with a duration of more than 20 days and a tilt velocity of more than 4.5ms/d without false prediction, but with one failure in earthquake prediction. For the strain CRD, 5 earthquakes occurred after the CRD; there was no false prediction, but 4 earthquakes unpredicted. We have tried to increase the thresholds, but found either the failure ratio or the rate of false prediction increased, while the success rate not obvious increased. Continuous Rapid Deformation (CRD) may be an important precursor before some strong earthquakes with magnitude greater than % M % s5.0. This paper shows some characteristics of CRD, and proposes a method for identifying precursory CRD that is closely related to the process of earthquake preparation. From the deformation data in Yunnan Sichan provinces, we obtained the tilt and strain thresholds of 4.5 ms / d and 5 × 10 -7 / d, respectively. Compared the tilt CRD in this region before 9 earthquakes ( short term fore shocks and after shocks not included) with magnitude greater than 5.8 in the period from January of 1990 to May of 2001, we found that 8 earthquakes occurred within 5 months after the CRD with a duration of more than 20 days and a tilt For the strain CRD, 5 earthquakes occurred after the CRD; there was no false prediction, but 4 earthquakes unpredicted. We have tried to increase the thresholds, but found either the failure ratio or the rate of false prediction increased, while the success rate not obvious increased.