This paper presents a new physics-based statistical model for random telegraph noise in Flash memories. From the probabilistic superposition of elementary Markov processes describing the trapping/detrapping events taking place in the cell tunnel oxide, the model can explain the main features of the random telegraph noise threshold-voltage instability. The results on the statistical distribution of the threshold-voltage difference between two subsequent read accesses show good agreement between measurements and model predictions, even considering the time drift of the distribution tails. Moreover, the model gives a detailed spectroscopic analysis of the oxide defects responsible for the random telegraph noise, allowing a spatial and energetic localization of the traps involved in the threshold-voltage instability process.