Abstract: To clarify the creep of rocks,it would be necessary to continue an experiment over several ten years. One of such experimental methods is introduced in this paper. A test-piece is shaped into a beam of about 20 cm length and its upper surface is highly polished. The test-piece is bent convex upwards; on that occasion one had better plan the maximum bending stress to be about 20 bar or less. While making a measurement, an optical flat is set above the polished surface to produce interference fringes of monochromatic light. From an analysis of the interference fringes, one can determine the profile of the upper surface.A bending of the test-piece should be reduced to a difference between the profile at the measurement and the initial profile which was determined before bending.Thus continuing the regular measurements of the bending curves of the test-piece, one could construct a creep curve. The necessary care to be taken and the problems encountered on applying the above method are mentioned,based on the author's experiences,and some concrete examples are shown.A bending test has a merit that detection of deformation is relatively easy and a demerit that a stress state is not simple as compared with a compression or tension test. In the last part, however, the author shows that the creep curve obtained by the bending test is comparable with the creep curve obtained by the compression or tension test, ifyield stress is zero. And he draws the important conclusion, from Kumagai and lto's creep experiment over 23 years, that rock does not have the yield stress.