Abstarct: Seismic imaging is highly dependent on the quality of seismic data. Structural and stratum interpretation of seismic sections that contain the least amount of coherent noise than random is much easier. There are many noises in seismic sections that random noises are one of the major categories. The attenuation of noise, especially when the signal-to-noise ratio is low, it is very difficult. So far several methods for random noise attenuation are introduced and each one has its own advantages and disadvantages. In this thesis we have utilized a new method for the attenuation of random noises in t-f-x domain and then applied EMD on each constsnt frequency slice. The first IMF obtained from EMD method, the data shows rapid oscillations. This means that the largest component of wave number (related to noise) is available on the first IMF. So to increase the signal-to-noise ratio, the first IMF should be subtracted from the original signal. Random noise consists of all frequencies, to attuanate the random noise in seismic sections, one should remove the frist IMF for all frequencies. The first IMF to remove the remaining noise in the other IMFs, the performance is not optimal. In cases where there are seismic events with steep dips, due to their high wave numbers, removing of the firat IMF will not perform optimally. To overcome this shortcome another new technique was used in this thesis. Insteade of removing frist IMFs we used thresholding interval. This new algorithm attenuations more noise than the first method and steep events remain unchanged. In this thesis the introduced algorithms applied on synthetic and real seismic sections and compared with a method baxsed on combination of Fourier Transform and EMD. Results indicate better performance of techniques presented in this thesis. It is also shown that removing of the first IMF in common shot gather can be efficiently attenuate ground roll as a coherent noise