Coherence analysis of the EEG is used to study the coupling between cortical regions. High coherence between signals recorded at 2 electrodes is interpreted as evidence for neuroanatomic connections between the cortical areas underlying the electrodes. When common reference recordings are used, coherence measures the relationship between 2 time series, each of which is the difference between 2 signals measured at the scalp and is confounded by spectral power and phase at the recording and reference electrodes. Using multi-channel EEG data from 3 subjects, we illustrate the confounding of common reference data coherence computations and demonstrate the extreme effects of power and phase changes on coherence by simulating these changes in the EEG data. Common reference coherence data can be either inflated or deflated as a consequence of activity (i.e., spectral power) at the reference. Phase relationships among the reference and recording time series modulate the power effects on coherence. Both the power and phase effects can vary dramatically across frequencies, having profound and complicated effects on the shape of the coherence function. Based on these considerations, we conclude that common reference coherence data must be interpreted very cautiously and recommend that a new body of EEG coherence data must be gathered using reference-free recording methods before the utility of EEG coherence analysis for understanding brain function can be determined.