Late potentials, low-level signals that outlast the normal QRS, are an index of VT risk. However, the reentry substrate for VT with which late potentials are associated may be active only within the normal QRS period. This paper introduces 2 linear modeling techniques for extracting intra-QRS late potentials. These are given a signal processing definition as the residual of an autoregressive model of the smooth, regular form of the QRS. They are a high resolution version of notches and slurs, QRS features associated with myocardial scarring. Intra-QRS late potentials were measured in a select group of patients assessed for VT. Preliminary results suggest they may have value as an index of VT risk.