A novel comprehensive system based on artificial neural networks (ANN) for detecting ventricular fibrillation (VF) using various signal descriptors is proposed. First, by using time-, frequency-domain, and nonlinear dynamics analyses, different kinds of descriptors with different meanings are extracted from cardiac rhythmic signals, and then incorporated into the detection system and make decision automatically via the comprehensive ability of ANN. The detection performance of the system is tested by actual ECG database. Finally, its feasibility in clinical application is discussed.