An in vitro experimental caries model is described in which an ultramicro system of analysis was used during simulated decay to study the composition of fluid in micro wells drilled within human enamel. In this experiment, hydrochloric acid was used as the demineralizing agent because the results could be directly compared to recent bench-scale studies using a two-compartment diffusion cell. The two sets of results agree in principle and generally confirm an earlier ‘diffusion-controlled’ caries model. Furthermore, the results are contrary to caries models based on the assumption that the rate-controlling process is release of ions from the enamel crystal surfaces because: (1) The solutions at various points in the lesion remained satured during the attack, although a highly demineralized lesion was used in the experiment, and a high degree of undersaturation was used to drive the demineralization. (2) The Ca/P ratio in the lesion solution increased during the demineralization. (3) The calcium and phosphate concentrations increased during the demineralization above their initial values although there was no concentration gradient for hydrochloric acid in this system. (4) Small membrane potentials developed during the demineralization, indicating that the tooth surface was permselective. (5) The directions of the changes in the concentrations of all the ions in solution could be predicted from the permselectivity using the ‘diffusion-controlled’ model. These results have major implications relative to the design of theoretical physicochemical models employed to elucidate the mechanism of caries.