In this paper, we study the magnetic flux leakage (MFL) signals for detection of defects in ferromagnetic materials. MFL signals consist of a background that is not constant and is combined with noise. Since there are slight variations because of noise, any large distortion shows a defect. Here the estimation of the background and then the determination of a threshold to distinguish defects from noise have been used for locating defects. In this method, precise evaluation of these two parameters has a vital role on the defect detection. The concept of histograms has been employed for eliminating the effect of defects in computing background signal. Results show that this algorithm is fast enough and yields detection of more defects that have lower amplitude. In the next step, an appropriate value for the threshold is determined by considering a trade-off between defect detection rate and noise separation rate.