Many algorithms, such as geospatial predicates or the construction of Delaunay Triangulations, make use of geometric predicates. Common examples of such predicates are orientation tests for 2D or 3D points. If these predicates are evaluated for floating-point coordinates, they can yield incorrect results due to round-off errors, which can cause algorithms to fail or to return invalid constructions. This talk presents a C++ template metaprogramming implementation of fast, robust floating-point predicates, which guarantee correct results without severe performance penalties.