It is well known that a compound layer may form at the target surface during reactive sputtering. However, the significance of this layer for the response to a change in target conditions has so far not been carefully investigated. The standard model for the reactive sputtering process [S. Berg et al., J. Vac. Sci. Technol. A 5, 202 (1987)] does not allow for calculations of the compound thickness at the target surface. For simplicity it has been assumed that a single monolayer is responsible for the poisoning of the target. However, experiments clearly indicate that the compound layer thickness may be significantly thicker than one monolayer. For several reasons it is important to be able to quantify the thickness of this layer. The formation of the compound layer introduces memory effects into the system when the processing conditions are changed. The sputter erosion time for the compound layer depends on its thickness. We will present an extension of the basic reactive sputtering model that allows for the formation of an arbitrary thickness of the compound layer at the target surface. From this model it is possible to examine the layer thickness dependence on the major processing parameters (reactive gas supply, target current, etc.). Experimental studies of the transient target voltage and process pressure support the validity of the new model.