Temperatures of the Arsia Mons caldera floor and two nearby control areas were obtained by the Mars Global Surveyor (MGS) Thermal Emission Spectrometer (TES). These observations revealed that the Arsia Mons caldera floor exhibits thermal behavior different from the surrounding Tharsis region when compared with thermal models. Our technique compares modeled and observed data to determine best fit values of thermal inertia, layer depth, and albedo. Best fit modeled values are accurate in the two control regions, but those in the Arsia Mons' caldera are consistently either up to 15 K warmer than afternoon observations, or have albedo values that are more than two standard deviations higher than the observed mean. Models of both homogeneous and layered (such as dust over bedrock) cases were compared, with layered-cases indicating a surface layer at least thick enough to insulate itself from diurnal effects of an underlying substrate material. Because best fit models of the caldera floor poorly match observations, it is likely that the caldera floor experiences some physical process not incorporated into our thermal model. Even on Mars, Arsia Mons is an extreme environment where CO2 condenses upon the caldera floor every night, diurnal temperatures range each day by a factor of nearly 2, and annual average atmospheric pressure is only around one millibar. Here, we explore several possibilities that may explain the poor modeled fits to caldera floor and conclude that temperature dependent thermal conductivity may cause thermal inertia to vary diurnally, and this effect may be exaggerated by presence of water-ice clouds, which occur frequently above Arsia Mons.