This paper presents a statistical characterization of random telegraph noise (RTN) in hafnium-oxide-based resistive random access memories (RRAMs) in high resistive state (HRS). Complex RTN signals are analyzed exploiting a Factorial Hidden Markov Model (FHMM) approach, which allows to derive the statistical properties of the RTN signals, directly related to the physical properties of the traps responsible for the multi-level RTN measured in these devices. Noise characteristics in different reset conditions are explored through consecutive switching cycles. Noise spectral analysis is also performed to fully support the investigation. An RRAM compact model is also exploited to estimate the physical properties of the conductive filament and of the dielectric barrier from simple I-V data. These tools are combined together to prove the existence of a direct statistical relation between the reset conditions, the volume of the dielectric barrier created during the reset operation and the average number of active traps contributing to the RTN. (C) 2014 Elsevier Ltd. All rights reserved.