The dip filtering based multiple reflections noise attenuation approaches depend on the separation between multiple reflections and primary reflections in the NMO corrected CMP gather, especially in the near offset where the local slopes between primary reflections and multiple reflections are very close. In this letter, we propose two novel strategies to better separate the primary reflections and multiple reflections. The first strategy is based on a trace randomization procedure, where we can turn the unflattened multiple reflections into random noise and maximize the difference between primary reflections and multiple reflections in the near offset. The next strategy is a different spatial smoothing approach. The new smoothing approach is based on the empirical mode decomposition (EMD), where we can attenuate the random noise that corresponds to multiple reflections adaptively, and more importantly we can preserve the useful reflection energy even though the events are not exactly flattened.