Oscillations threaten the stability of a power system. Timely detecting oscillations can improve operators' situational awareness of system stability and enable remedial reactions. To detect oscillations during their early stages, this paper proposes a cross-coherence method using multiple-channel phasor measurement unit (PMU) data. Oscillations are related to the peaks in coherence spectra and can be detected by visual inspection and setting up a threshold. Using simulation data, it is shown that the proposed coherence method can detect oscillations even under low signal-to-noise ratios. Three algorithms for estimating coherence spectra are implemented and evaluated. Their performances are compared using Monte Carlo methods. Based on the comparison, this paper makes some recommendations for proper use of the algorithms.