The invention provides a 3D earthquake data de-noising method based on 2D curvelet transform. The method is characterized in that a time slice of the noise-contained 3D earthquake data is extracted, and multi-dimension multi-direction 2D curvelet transform is carried out on the time slice to obtain a curvelet-domain coefficient; a local threshold de-noising method is used in the curvelet domain, one soft threshold operator is selected for each dimension after curvelet transform, and effective-wave curvelet coefficients in different dimensions are obtained via threshold processing; and the extracted effective-wave curvelet coefficients are used to carry out inverse transformation to reconstruct earthquake signals, and the aim of de-noising is achieved. According to the invention, 2D curvelet transform is used for 3D earthquake data de-noising, the local threshold de-noising method is provided, the soft threshold operators are used, each time slice is de-noised independently, the disadvantages that a traditional global threshold method is not ideal in the de-noising effect, requirement for memory of a computer is lowered, the computing efficiency is greatly improved, and operation time is reduced.