Ground-roll, characterized by low frequency and high amplitude, is one of the most important issues in seismic data processing. Common methods to eliminate ground-roll noise from seismic data include frequency-based methods which pose filtering the whole signal. In addition, assuming being stationary signal which attended in filter is another disadvantage of those methods: however, in this paper a new method based on using Hilbert transform to address high-amplitude, low-frequency coherent noise is proposed. In this respect the conventional seismic trace can be viewed as the real component of a complex trace. Therefore, its envelope and normalized phase, as the input of running average, are easily calculated. On the other hand, complex trace-based filtering is approached to be a method distinguishing between coherent noise and reflections overlapping each other. This is done by setting a suitable length of window on running average in order to extract low frequency components from data. (C) 2014 Elsevier B.V. All rights reserved.