As the interconnection of power grids, failure of single component could propagate out, and because of the scale of the power system, it is almost impossible to analysis all failure patterns. Traditional methods for online cascading analysis could face almost unsolvable obstacles on computational burden and accuracy. This paper discusses the mechanisms of early warning of cascading blackouts. Researches have shown, from the complex system point of view, the existence of early-warning signals to indicate for a wide class of systems whether a critical threshold is approaching. Instead of focusing on identifying the possible failure patterns that could lead to cascading blackouts, from the complex system point of view, early warning of cascading blackouts could be realized by identifying how far the system is to the critical state by simulation using fast computational techniques and a group of system indices, providing helpful aid for the operator.