This paper considers how data based neurofuzzy modelling techniques for the poorly understood relationships between changing process histories and the evolution of the internal state variables of dislocation density, subgrain size and subgrain boundary misotientation can be combined with physically-based models to investigate the effects of the internal state variables on the flow stress and recrystallisation behaviour. The model uses genetic algorithms to optimise the constants and is validated for data on a range of aluminium-magnesium alloys of both high and commercial purity. It is shown that this hybrid modelling methodology supported by a knowledge base offers a flexible way to develop the microstructrural modelling as more data and better understanding of the evolution of the internal state variables become available.