During an earthquake, many stations record the ground motion, but only a few numbers of them could be corrected using conventional high-pass and low-pass filtering method and the others identified as highly contaminated by noise and as a result useless. There are two major problems associated to these noisy records. First, since signal to noise ratio (S/N) is low, it is not possible to discriminate between original signal and noise neither in frequency domain nor in time domain. The consequence is that it is not possible to cancel out the noise using conventional filtering methods. The second problem is about non-stationary characteristics of the noise. In other words, in many cases the characteristics of the noise are varying by time and in these situations, it is not possible to apply frequency domain correction schemes. For correction of acceleration signals contaminated with high-level non-stationary noise, there is an important question that is it possible to have estimation about the state of the noise in different bands of time and frequency? The wavelet multi-resolution analysis which decomposes a signal into different time-frequency components besides a suitable criterion for identification of the noise among each component provides the required mathematical tool for correction of highly noisy acceleration records. In this paper, the strong motion records of great earthquake events with Mw>5.5 in Iranian Plateau is corrected. For this purpose, first, signal to noise ratio is calculated for each record. If the value of this diagram is high enough, the signal is corrected with both methods and otherwise, the de-noising is applied using wavelet procedure. The effect of baseline correction of the final results is also investigated for each record. A comprehensive database containing the information of events, stations and correction parameters is also provided in this study.