This study performed equilibrium experiments in order to evaluate previously proposed formulations of the garnet (Grt)-clinopyroxene (Cpx) thermometer as applied to eclogites. The starting material is fine-grained powder of natural eclogite (< 10 mu m), whose main constituents are Grt (Fe:Mg:Ca similar to 44:28:28), Cpx (Na pfu similar to 0.55-0.60), phengite, quartz and rutile. Experimental conditions are 1,100-1,250 degrees C at 2.5 GPa, and the run duration is 193-334 h. The experimental run products mainly consist of Grt, Cpx, and glass. In a preliminary experiment at 1,000 degrees C for 144 h, Cpx grains are clearly zoned and most Grt grains maintain primary compositions. In the higher T (>= 1,100 degrees C) and longer run (>= 193 h) experiments, Cpx in the run products becomes poorer in Na and higher in Fe/Mg compared with the starting material, and each grain does not show clear chemical zoning. Garnet compositions become poorer in Ca [Ca/(Fe+Mn+Mg+Ca)similar to 0.2-0.25] and lower in Fe/Mg compared with the starting material. The average composition of Cpx and the average of Ca-poor Grt compositions in each run product were used to evaluate previously proposed formulations of the Grt-Cpx thermometer. Temperatures calculated with formulations by Pattison and Newton (1989) and Berman et al. (1995) are much lower than the experimental temperatures, even though these formulations are based on the compositional bracketing-type experiment. One of the reasons for this discrepancy might be uncertainty of solid-solution properties of Al in Cpx, because the value of the excess interaction parameter for Al in the generally low-Al Cpx modeled by Berman et al. (1995) is much higher than those proposed by independent experiments, resulting in the estimated temperatures being significantly lower than the experimental temperatures.