This research involved a third and final phase of the intelligent image-based winter road condition sensor research effort. From the beginning, the goal of this third phase was to use one road condition classification system on all camera sites in Sweden. When researchers discovered in this phase that the accuracy was far too low when the equipment was moved to new sites, it was soon realized that the potential benefit disappeared immediately. The first two phases of the project had shown significant promise, and this third phase continued research and movement of the test site to new locations in order to acquire additional research data. Three neural networks were used in the field tests: one for day, one for night, and one combined day/night network. Initial tests were done in Matlab to get information about the performance of the three different neural networks. Test one showed that the classification was very correct even within a reasonable confidence value. It was also noted that in the current situation, only two types of road conditions, dry and snow, were being classified. In the 2005-2006 winter, up to five classes of road conditions were able to be detected, and a second camera was set up to verify the accuracy of the first. Unfortunately, as noted above, the field image classification system was far too low in accuracy to be acceptable.