The use of a hadron collider, like CERN's LHC, as a research tool at the high energy fronteer makes substantial demands on the theoretical understanding and predictive power of QCD.
In the past few years, severe efforts have been made to extend the predicitveness of
perturbative QCD by computing the next-to-next-to-leading order (NNLO) corrections to the
production rates of several basic processes. In the talk I will review the motivations for
NNLO computations and highlight some of the recent developments. In particular I will discuss progress towards setting up a universal algorithm for computing QCD cross sections at NNLO accuracy based on the subtraction method.