The residual dentin thickness in the coronal part of root canals of mandibular premolars after stepback preparation and flaring with Gates Glidden (GG) drills was assessed sequentially in a newly designed muffle device. Twelve extracted teeth were embedded in clear polyester resin, sectioned horizontally (1, 3, and 5 mm apically to the cementoenamel junction), and reassembled in the muffle device. The sequence of canal preparation was K-files to #40, then GG-2 and GG-4. After each procedure, the slices were separated and residual dentin thickness measured by a toolmaker microscope in four directions (buccal, lingual, distal, and mesial) and reassembled in the muffle device. Statistical analysis by three-way analysis of variance with repeated measures was undertaken. The difference in residual dentin thickness was highly significant with regard to instrument (control, K-40, GG-2, GG-4; p < 0.0001), slice (upper, middle, and lower; p < 0.0003), and direction (B, L, M, and D; p < 0.001). In each slice, the width of the mesial side was similar to the distal side, as was the buccal to the lingual sides. Reduction of residual dentin thickness in the mesiodistal direction, from the unprepared upper slice to the GG-4 prepared lower slice, was appreciably greater (35%) than in the buccolingual direction (5%).