An algorithm to predict crustal thermoelastic strain from observed local atmospheric temperature is given and applied to a 24-month crustal strain record of one test strainmeter site located near Bouquet Reservoir in southern California. We use a crustal model that consists of an elastically decoupled layer overlying a uniform elastic half-space, and a thermal source that is given by a stationary temperature wave whose wavelength is related to local topography and/or lateral material heterogeneity. The decoupled layer delays, attenuates, and low-pass filters the source temperature field. The thermoelastic strain in the underlying half-space, resulting from the temperature variations at the base of the decoupled layer, is calculated using the Berger (1975) solution for thermoelastic strain in a uniform half-space. Applying our model to the test data, we obtain a good fit between predicted and observed strains if we filter the surface thermal signal through a 63-cm-thick decoupled layer. Much of the remaining strain variations clearly correspond to other environmental sources (reservoir loading and rainfall). Our analysis suggests that the horizontal thermoelastic strain is inversely proportional to the wavelength of local topography and/or lateral material heterogeneity. Thus, the horizontal thermoelastic strain will be greater in areas of local topography and/or lateral material heterogeneity and smaller in more homogeneous and flat areas. An upper layer of loose material, natural or artificial, acts as a thermoelastic strain insulator. Burial of strainmeters in places where such a layer exists can reduce the thermoelastic strain noise considerably even for shallow strainmeter emplacements.