Volume conduction is a major problem in spatial-temporal electroencephalogram (EEG) signals. We propose a frequency dependent subtraction approach to address the volume conduction problem in EEG signals. We validate the proposed approach using simulated data and discuss the application of the approach to the EEG data recorded from three sick infants. The frequency dependent subtraction mitigated the common signal better than the traditionally used re-referencing the EEG signals to the global average of EEG from all electrodes. Furthermore, the frequency-based approach also attenuated the other interfering signals such as noise from power line and mechanical ventilators used to support patients.