During plastic deformation, most of the dissipated work is released as heat, but a fraction of it, usually small, is stored in the microstructure, is called latent heat and is associated with the network of dislocations that develops. The rate of energy storage in the microstructure divided by the rate of plastic dissipation is defined as the latent heat capacity. Latent heat remains stored in the microstructure of cold worked specimens after quenching. This energy is associated with modified mechanical properties, e.g. hardness, and is released upon annealing. Saturation of this stored energy has been observed in experiments after a specific amount of plastic deformation is reached. A thermodynamically consistent model for continuous dynamic recrystallization is proposed in this paper with the aim of explaining the phenomenon of latent heat saturation and relating it to grain refinement. The proposed model has three essential features: (i) the latent heat increases in the specimen during plastic deformation as plastic work is continuously dissipated; (ii) the rate of latent heat storage per unit work, i.e. the latent heat capacity, is related to the internal architecture of the microstructure and decreases to zero as a consequence of microstructural evolution; (iii) the relationship between the latent heat and the microstructure is described through the use of two parameters: (a) the dislocation density and (b) the average grain diameter. A comparison of the proposed model with experiments is reported and a validation for the prediction of microstructural evolution, as well as the evolution of the latent heat and latent heat capacity, is provided.