Early detection of life-threatening arrhythmias such as ventricular fibrillation (VF) and rapid ventricular tachycardia (VT) is most essential for an automatic external defibrillator and remote cardiac patient monitoring. In this paper, we present a low-complexity, robust detection method for automatically detecting VT and VF events in the ECG signal. The proposed detection method consists of four stages: (i) discrete Fourier transform (DFT) based noise removal, (ii) addition of random noise; (iii) zero-crossing rate (ZCR) estimator and (iv) detection rule. The proposed method is tested and validated using the MIT-BIH arrhythmia database (MITADB), Creighton university ventricular tachycardia database (CUDB) and the MIT-BIH malignant ventricular arrhythmia Database (MITMVFDB). The method achieves an overall sensitivity (Se) of 99.34%, and specificity (Sp) of 99.97% for 5 s signal. The robustness of the method is tested using different types of PQRST morphological patterns and various kinds of noises including the baseline wanders, powerline interference, muscle artifacts and white Gaussian noise. The results demonstrate the proposed method with additive random noise with single ZCR feature can achieve significantly better detection rates as compared with the existing detection methods based on the combination of morphological, spectral, time-frequency, and complexity features and the machine learning techniques such as neural networks, support vector machines (SVM), fuzzy neural networks (FNN). © 2016 IEEE.