Bi-frequency Coherence is a normalized spectrum function based on Loeve Spectrum. In this work, we investigated the use of Bi-frequency Coherence function for correlation of two non-stationary processes, particularly for EEG signals. We define a procedure to achieve Bi-frequency Coherence starting from Loeve Spectrum. Then we present a numerical estimation of this approach. We compare our method with traditional coherence and TF-coherence functions, by means of some examples to show its advantages. It is shown that for some non-stationary processes, the proposed Bi-frequency Coherence function may extract underlying coupling better than other approaches.