The recent development and increased application of automatic external defibrillators have prescribed very strong requirements towards ventricular fibrillation (VF) and fast ventricular tachycardia (VT > 180 bpm) detection from the surface electrocardiogram (ECG). We attempted to use informative parameters from several existing analysis methods and from a method developed in-house. A set of nine parameters was derived initially, with four of them being selected after statistical assessment. Detection of VF against non-shockable rhythms was obtained using the K-nearest neighbours classification method, with 98.6% specificity and 96.7% sensitivity. The detection accuracy remained high after inclusion of VT episodes above and below 180 bpm to shockable and non-shockable rhythms respectively and after the addition of noise. Test signals were taken from the well-known ECG signal databases of the American Heart Association and the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH-'cudb' and 'vfdb' files).