Caries mainly occurs at retention sites for plaque such as pits, fissures, margins of restorations and interproximal sites. An in vitro model was developed to investigate dentine caries in narrow gaps. Forty bovine coronal dentine discs were covered with bonding agent. In 20 discs, two grooves, 200 and 340 mu m wide and 500 mu m deep, were sawn to mimic fissures. In the 20 other discs, a 1-mm-wide groove was made and, after polyester sheets were placed against both walls, filled with composite. After curing of the composite and removal of the sheets, grooves of 22 mu m remained at the borders of the restoration. All grooves were exposed to 8% methylcellulose gel 0.1 M lactic acid at pH 4.8 for 1 week. Demineralization was determined by microradiography of sections sawn out of the center of the discs. The walls of the grooves showed subsurface lesions, which decreased in size towards the base of the grooves. The average mineral losses (vol % x mu m) at the entrance of the 22-, 200- and 340-mu m-wide grooves were 1,112 (SD 370), 1,277 (293) and 1,277 (255), halfway down the groove 218 (150), 659 (244) and 797 (207) and at the base of the groove 140 (88), 285 (145) and 504 (205), respectively. Analysis of variance and Tukey's B test showed that the average mineral losses from grooves of different width and from sites at different depth were all significantly different at p<0.05 level, but not for the mineral losses from the entrance of the two wider grooves and from the two lower sites in the narrowest groove. The reduced mineral loss in the two narrower groove widths is assumed to be caused by the more limited inward diffusion of acids and outward diffusion of mineral ions through the methylcellulose gel.