From experimental evidence it is well known that the bioavailability of controlled release systems, i.e., the percentage of the dose absorbed by the body, is often reduced compared to a corresponding dosage form with immediate release. In the case of inert matrices, a water-soluble drug is embedded in a finely dispersed state in an insoluble carrier material and released by diffusion. In the present work such systems are described by percolation theory. Based on a Bethe lattice model the amount of drug substance trapped in the matrices, which determines the reduction of bioavailability, is calculated in a straight-forward way from the volume-to-volume ratio of drug and matrix material. To check the use of the model, matrix tablets are prepared with caffeine as a model drug and ethyl cellulose or hydrogenated castor oil as carrier materials, and their drug release is determined in vitro. The experimental findings are in good agreement with the values predicted from the percolation model. The most pronounced reductions of bioavailability are observed if the volume-to-volume ratio of drug and matrix substance is below a percolation threshold.