The common average reference (CAR) is influenced strongly by high-amplitude artifacts at the recording electrodes. The authors suggest an algorithm for average reference modification that allows application of the average reference montage in the presence of high-amplitude artifacts at one or several EEG data channels. This dynamic average reference (DAR) method is based on analysis of kurtosis (DARkurt) or kurtosis and standard deviation (DARSD) of amplitude distribution at a single time point and elimination of outliers before calculating the average reference. This procedure can be performed on-line. The DAR improves visual EEG representation considerably in the presence of high-amplitude artifacts. The DAR also influences artifact-free EEG, increasing slightly the amplitude of high-amplitude waveforms. The modeling of the dipole source in the spherical conductor has shown that the DAR compared with the CAR can both decrease and increase the average reference bias depending on position and orientation of the source. In practice, the influence of the DAR with tested parameters on the artifact-free EEG was low and never exceeded 12% of the amplitude of the CAR-referenced signal. The DAR method can be useful for routine analysis of EEG contaminated by high-amplitude artifacts and for long-term EEG monitoring.