A major criterion for assessing the value of any experimental model in scientific research is the degree of correspondence between its results and data from the real-life process it is designed to model. Intra-oral models aimed at predicting the anti-caries efficacy of toothpastes or other topical treatments should therefore be calibrated against treatments proven to be effective in a caries clinical trial. For this to be achieved, it is necessary that a model with high sensitivity be designed, while at the same time retaining relevance to the process to be modeled. This means that the effects of the various experimental conditions and parameters of the model on its performance must be understood. The purpose of this paper was to assess the influence of two specific factors on the performance of an in situ enamel remineralization model, which is based on human enamel slabs attached to partial dentures. The two factors are initial lesion severity and origin of enamel sample. The results indicated that initial lesion size affected whether net remineralization or net demineralization occurred during in situ treatment. Samples with an initial range of from 1500 to 2500 (delta Z) tended more toward demineralization than did samples with delta Z greater than 3500. This means that treatment groups must be well-balanced with respect to initial lesion size. Differences in initial demineralization severity between different tooth locations must also be considered so that systematic treatment bias can be avoided. The solution used in the model discussed here is based on a balanced experimental design, which allows this effect to be taken into account in the data analysis.