Objective: According to previous studies, the complex substrate of malignant arrhythmias needs a detailed spatio-temporal noninvasive characterization of low-amplitude dynamic changes in beat-to-beat cardiac repolarization. Methods: Body surface potential map (BSPM) records were taken on 14 healthy male and female subjects (age 20-80 years) and on 6 ventricular arrhythmia patients, 4 of them with implanted cardioverter defibrillators (ICD). Records were taken continuously, for 5 minutes, in resting, supine position. Beat-to-beat QRST integral maps, Karhunen-Loève coefficient time-series (KLi, i=1-12), RR and nondipolarity index (NDI) time-series were computed. Results: The first order statistical properties of the spatio-temporal variability of subsequent QRST integral maps were characterized by the box and whiskers plot of their KLi components. The SD2/M2 (KLi amplitude variance/mean signal energy) values of the QRST integral maps in the normal group ranged between 0.0057 and 0.008 (i.e. 0.075£SD/M£0.089). In ICD patients SD2/M2 values went up to 0.021-0.069 (i.e: 0.14£SD/M£0.26). Autocorrelation functions revealed that while in normal subjects only 5-20% of the total power had white noise character, the rest was bandwidth-limited noise. In ICD patients the weight of white noise component increased considerably. The higher the SD/M relative KL variability, the higher and more frequent NDI spikes were. Conclusions: Beat-to-beat dynamics of white noise components of high resolution BSPMs are able to stratify arrhythmia vulnerability. The temporal distribution of extreme NDI spike formations is random; the frequency is associated with the relative KL component noise levels.