We introduce two ways to produce locally calibrated grid-based probabilistic forecasts of temperature. Both start from the Bayesian model averaging (BMA) statistical post-processing method, which can be globally calibrated, and modify it so as to make it local. The first method, geostatistical model averaging (GMA), computes the predictive bias and variance at observation stations and interpolates them using a geostatistical model. The second approach, Local BMA, estimates the parameters of BMA at a grid point from stations that are close to the grid point and similar to it in elevation and land use. We give results of these two methods applied to the eight-member University of Washington Mesoscale Ensemble (UWME) for the 2006 calendar year. GMA was calibrated and sharper than Global BMA, which has constant predictive bias and variance across the domain, with prediction intervals that were 8% narrower on average. Examples using a sparse and dense training network of stations are shown. The sparse network illustrates the ability of GMA to draw information from the entire training network, while Local BMA performs well in the dense training network due to the availability of nearby stations that are similar to the grid point of interest.