A quantitative technique is presented for discrimination of ventricular fibrillation (VF) signals from other mono-morphic and polymorphic ventricular arrhythmias. Ventricular arrhythmias are analyzed by means of a two-step procedure: (i) a short-term autocorrelation function (SACF) is calculated, and (ii) a regression test is carried out on a plot of peak magnitudes of ACF. A plot of SACF lag vs. peak order (SACF-lag plot) should be linear for ventricular tachycardias, but not for VF. This hypothesis was tested on 31 episodes of ventricular arrhythmias recorded from humans during electrophysiologic procedures. SACF-lag plots were during for three successive 1. 5-s segments of each record and tested for a regression fit (by an F-test) to the ACF-lag plot. The VF discrimination algorithm was found to perform well after only one segment and to improve its performance until, after the third segment, sensitivity, specificity, and test efficiency were all 100%. This technique seems applicable for accurate discrimination of ventricular arrhythmias.