The diurnal temperature cycle in the Phoenix metropolitan area, as represented in the PSU/NCAR model Version 5 (MM5) is examined using high resolution (2 km grid spacing) during the dry portion of the summer. The model is run for a 48 hour period with negligible synoptic forcing and a sufficiently dry atmosphere that moist convection does not occur. The standard version of MM5 underestimates the magnitude of the diurnal maximum and also causes the maximum to occur too early. The deficiency in the temperature forecast is linked with an overestimate of the latent heat flux in the model, which results from a poor specification of the land cover category and corresponding physical parameters. Adjusting the available moisture improves the forecast temperature maximum at both rural and urban sites, but the model temperatures are still cooler than observed at the urban site at night due to a poorly represented urban heat island effect.