Modern hearing aids use Dynamic Range Compression (DRC) as the primary solution to combat Hearing Loss (HL).Unfortunately, common DRC based solutions to hearing loss are not directly based on a proper mathematical or algorithmic description of the hearing loss problem. In this paper, we propose a probabilistic model for describing hearing loss, and we use Bayesian inference for deriving optimal HL compensation algorithms. We will show that, for a simple specific generative HL model, the inferred HL compensation algorithm corresponds to the classic DRC solution. An advantage to our approach is that it is readily extensible to more complex hearing loss models, which by automated Bayesian inference would yield complex yet optimal hearing loss compensation algorithms.