This paper proposes a multi-layer interaction graph on cascading outages of power systems as an extension of a single-layer interaction network proposed previously. This multi-layer interaction graph provides a practical framework for the prediction of outage propagation and decision making on mitigation actions. It has multiple layers to, respectively, identify key intra-layer links and components within each layer and key inter-layer links and components between layers, which contribute the most to outage propagation. Each layer focuses on one of several aspects that are critical for system operators’ decision support, such as the number of line outages, the amount of load shedding, and the electrical distance of outage propagation. Besides, the proposed integrated mitigation strategies can limit the propagation of cascading outages by weakening key intra-layer links. All layers are constructed offline from a database of simulated cascades and then used online. A three-layer interaction graph is presented in detail and demonstrated on the Northeastern Power Coordinating Council 48-machine 140-bus system. The key intra- and inter-layer links and key components revealed by the multi-layer interaction graph provide useful insights on the mechanism and mitigation of cascading outages, which cannot be obtained from any single-layer.