This paper proposes a novel islanding fault detection scheme based on the networked ellipsoidal estimation method for distributed grid-connected photovoltaic (PV) generation systems. The detection of the islanding condition is determined by computing the intersection of the global ellipsoidal estimations. While the global estimations are integrated at each sampling instant with all the local ellipsoidal estimations provided respectively by the local estimators in the communication network. If the islanding fault exists, the intersection of the global ellipsoids would be empty. If the islanding condition does not happen, the intersection of the global ellipsoids would be non-empty. The conditions on the existence of the local and the global ellipsoidal estimation methods are derived by two convex optimization approaches respectively. A simulation is provided by Matlab based on a model of 2-kW single-phase grid-connected power generation system to illustrate the effectiveness of the designed scheme on the detection of the islanding fault.