Abstract: Fortherandomnoisesuppressinginseismicrecordswithlowsignaltonoiseratio， traditionalmethodswill
harm the signal components. The paper thinks curvelet transform can separate the random noise using multi-scale
and multi-direction. The authors apply Monte Carlo estimator to compute the noise level and design a non-linear
thresholding function to remove the random noise coefficients， so the useful signal will be recovered. Applications
on both synthetic data and actual seismic data from Dabashan area show that the new method eliminates the noise
portion of the signal more efficiently and retains a greater amount of geologic data. The quality and consecutive of
seismic event are better as well as the quality of section is improved obviously， and it overcomes the drawback that
the conventional filtering approach may affect the effective wave when suppressing noise.

Key words: Curvelet, Monte Carlo, non-linear thresholding, random noise, Dabashan area