In the absence of a vaccine for use in humans against West Nile virus (WNV), mosquito control and personal protection against mosquito bites are the only measures available to prevent disease. Improved spatial targeting is desirable for costly mosquito and WNV surveillance and control schemes. We used a multivariate regression modeling approach to develop spatial models predicting high risk of exposure to WNV in western and eastern Colorado based on associations between Geographic Information System-derived environmental data and zip code of residence for 3,659 human WNV disease cases from 2002 to 2006. Models were robust, with user accuracies for correct classification of high risk areas of 67-80%. The importance of selecting a suitable model development area in an ecologically and climatically diverse environment was shown by models based on data from the eastern plains landscape performing poorly in the mountainous western part of Colorado and vice versa.