This work presents the development of a new probabilistic model for describing the residual stress state in hot-rolled I-sections. After addressing the limitations of previous deterministic and probabilistic approaches, three different types of stress distribution patterns are evaluated. These patterns are fitted to a dataset of sixteen measurements compiled from the literature, and their performance is assessed based on the deviations observed. In doing so, it is also demonstrated that the commonly used ECCS model exhibits significant shortcomings. Finally, pattern-related stochastic parameters are determined, providing a modelling framework for more precise structural analyses.