This paper develops a theoretical and methodological framework that integrates Hedonic Pricing (HP), grid comparable sales approach (CSA), and nearest neighbors into a general spatiotemporal specification. By explicitly providing a theoretical justification for introducing spatial (or spatiotemporal) econometrics to HP, this approach is not only relevant to house price forecasting and automated valuation models (AVM) but also to valuing environmental goods capitalized in housing and to all other fields employing house pricing models. The resulting econometric CSA and a spatiotemporal Durbin model provide higher prediction accuracy to alternatives and minimize the spatially-delineated omitted variable bias (OVB) common in HP. Spatiotemporal autoregressive (STAR) and error models (STEM) are also derived, providing specific conditions, under which their application can be justified. Our analysis reinforces the common real estate practice of selecting a small number of comparables in grid CSA and challenges AVM approaches, in which hundreds or thousands of comparables are introduced.