Gold is used widely for capsules in high-temperature rock-melting studies because it is generally thought to absorb negligible Fe from silicate samples. However, we observed significant losses of Fe from fluid-absent melting experiments on hornblende gabbros at 800-975 ??C and 8 kbar, using standard piston-cylinder techniques. The extent of Fe loss from the sample is dependent on the relative masses of the sample and the capsule. Low sample to capsule mass ratios (~0.04) lead to the highest Fe losses (32-49% relative). Concentrations of Fe in silicate melt and used gold capsules define an apparent equilibrium constant (K') that follows a linear 1n K' vs. 1/T relation (at an estimated log f(O)(2) of QFM-1). The apparent equilibrium constant is used to make limiting upper estimates on the amount of Fe that could be lost during rock-melting experiments for a range of f(O)(2) and sample to capsule mass ratios. At high f(O)(2) (NNO + 2), loss of Fe to gold is negligible (<2% relative) for a wide range of sample to capsule mass ratios. At an f(O)(2) of NNO, Fe loss can be kept to <10% relative by using a sample to capsule mass ratio of 0.2 or greater. At low f(O)(2) (QFM-1), presaturating the Au with Fe would be necessary to ensure that Fe losses remained <10% relative. Fe loss can compromise experimental results for small samples run at low f(O)(2) conditions, be they buffered, imposed by the pressure media, or produced by intrinsically reduced (graphitic) starting materials.