Surface temperature of alkali halide is measured by a new extrapolation method. The sample surface is KBr(001), which is fixed on a sample heater by a flat spring in a vacuum chamber. Preparing small pieces of KBr other than the sample, copper-constantan thermocouples are inserted into three points A, B, and C, and temperatures, TA, TB and TC, are measured. Point A is between the heater and the KBr sample, B is between the KBr and a small piece of KBr on the sample, and C is between the KBr piece and tip of the flat spring. Temperatures TB and TC are measured for various thickness of KBr pieces without changing their cross-sectional area. The temperature difference between TB and TC is expected to decrease with decreasing thickness of the KBr piece, because both temperatures would approach to a unique value for negligibly thin piece. Extrapolating the temperatures TB and TC to zero thickness, we predict the surface temperature, TS. The temperatures TS are obtained for several heater temperatures, and temperature rise, "TS, by the heating are 70 to 80% with respect to those of temperatures TA. They are compared with TB extrapolated to zero thickness, TB 0. Disagreement between TS and TB 0 becomes obvious with increasing temperature. Thus some techniques for measurements are reviewed and a simple calculation for conduction of heat is examined. Consequently, this disagreement is attributed to cooling of the flat spring by radiation and thermal contact resistance of existing gap at the interface between the sample and piece.