Abstract Most of the models used for taking into account the influence of mean stress on fatigue life work on the basis of the correction of stress amplitudes of particular loading cycles. This can be done directly on the amplitudes or through the modification of the reference S–N curve. The mentioned procedure is also applicable for variable amplitude loading where, after rainflow cycle counting, all loading cycles described by amplitude and mean-stress value are designated. Mean-stress correction can be difficult to apply in stochastic fatigue-damage accumulation, also called the spectral method, where the load and the stress level is defined in the frequency domain using the appropriate power spectral density (PSD) function. The PSD function by itself does not give information to the user about the level of mean stress and only provides us with some statistical information of the fluctuating part of the random loading processes. The aim of the paper is to present how we can perform the mean-stress effect correction directly on a PSD function of stress using well-known models, such as proposed by Gerber, Goodman, Soderberg, or Morrow. The correction can be performed depending on the frequency component, which is a significant advantage of the presented method. The theoretical elaborations were used to develop a computer simulation used to compare the frequency-domain with the time-domain computation path. To show the effectiveness of the presented method, fatigue life has been estimated and compared with experimental fatigue test results for S355JR steel.