At the heart of most training and analysis systems used by the warfighter is a consistent, realistic, and valid visual and topologic representation of the terrain; visual terrain features such as rivers and roads; and visual models of vehicles, trees, and buildings. Although the industry has made significant advances in standardization, multiple simulation systems when interoperating still require multiple runtime terrain formats. Even without the interoperability use case, the simulation engineering community is often faced with datasets of the same terrain in many different formats. Achieving sufficient correlation between these multiple representations of the terrain is a necessary condition to support the warfighter, but just what is sufficient? Most existing methods of terrain correlation rely on imperfect assumptions, are manpower-intensive and time-consuming (and thus error-prone), and are even somewhat ad hoc. Using advances in processing power, especially those in graphics processing unit (GPU) technology, we examine in detail a range of possible terrain correlation problems in both elevation and line of sight (LOS), two key measures of correlation. This paper explains both the basic methodology behind this advanced terrain correlation testing and summarizes quantitative results. Using these techniques, our detailed examination of very large amounts of data in multiple datasets reveals potentially significant and hitherto uncovered problems in terrain correlation. A more detailed understanding of these problems is expected to provide insight into the quality of existing databases, the impact on training effectiveness of inadequately tested terrain databases, and even how very large datasets can be compared for other correlation problems.