A hydrothermal flow-through experiment was performed for an artificially created single tensile fracture of granite. Water that had dissolved granite (Si concentration: 250 ppm) was injected into the fracture under hydrothermal conditions (effective normal stress: 10-13 MPa, temperature: 150°C) during 450 hours. Non-monotonic changes of fracture permeability were observed. Fracture permeability decreased significantly only during first 150 hours. Si concentration of water produced from the fracture was increased monotonically. However, the Si concentration was smaller than that of water injected into the fracture. A numerical modeling using experimental data was also performed for hydrothermal flow in the fracture. Aperture structures for the beginning and the end of the experiment and resulting hydrothermal flow were determined using experimentally obtained fracture surface geometries and fracture permeability. For the beginning of the experiment, developments of preferential flow paths (channeling flow) in heterogeneous aperture structure were observed obviously, while the developments of preferential flow paths were not observed obviously for the end of the experiment. It was also observed that local apertures around the preferential flow paths for the beginning of the experiment tended to become smaller at the end of the experiment, while other local apertures for the beginning of the experiment tended to become greater at the end of the experiment. Although both increase and decrease of local apertures were observed between the beginning and the end of the experiment, arithmetic mean values of local apertures for the end of the experiment (71 μm) was smaller than that for the beginning of the experiment (84 μm).