This paper proposes a novel method for diagnosing the sources and propagation paths of plant-wide oscillations. The main oscillatory process loop variables and dominant oscillation frequency of the process loop variables were automatically selected based on a spectral independent component analysis algorithm. The complex causality among the selected oscillatory process variables was then revealed by partial directed coherence whose magnitude at the dominant oscillation frequency represented the intensity of the influence among the process variables and was visualized in the form of a causality diagram. The location of the oscillation disturbance was identified intuitively through the final causality diagram. Simulations and an industrial case study demonstrated the effectiveness and applicability of the proposed procedure.