Repolarization alternans (RA), originating at the cellular level, is thought to produce an arrhythmogenic substrate, and surface ECG T-wave alternans (TWA) is a marker of risk for sudden cardiac death. In this paper we study RA in the unipolar electrograms (EGM), which represent the electrical activity of the heart at the tissue level. We first describe a simple analytical model to study how RA, simulated as alternating variations of action potential duration, affects EGM-TWA, and then we propose a novel methodology based on time-frequency analysis to detect EGM-TWA which occurs intermittently in few consecutive beats. In a simulation study, we used a 257-node whole heart model to reproduce several patterns of RA. RA involved specific subsets of adjacent nodes (11, 65 and 257), exhibited different amplitudes (0.25, 0.5 and 1 ms) and lasted for 40 consecutive beats of a 80-beat-long test sequence. Results show a relationship between the spatial distribution of RA and EGM-TWA: the smaller the region where RA occurs, the higher the extent of EGM-TWA. With the proposed methodology, we localized those portions of myocardium which exhibited EGM-TWA with an accuracy higher than 90%.