The aim of this paper is to compare various real estate valuation models and the manner in which they take into account environmental variables. The reference model is taken to be a standard linear regression model including ordinal variables to measure environmental quality. This type of model is widely used. It is first compared to linear models which incorporate environmental quality notes extracted from the urban habitat database of a Geographic Information System (GIS) which has been developed recently for Geneva, Switzerland. We also incorporate these quality notes in a single input parameter, a so-called geo-index. The price indices constructed in this way are quite similar to the more traditional hedonic model. We additionally find that Artificial Neural Network (ANN) models, which are non-linear per se, exhibit a similar general form of the price indices, but that the detailed price behaviours of different models feature notable differences depending on the input choice of environmental variables.