Molecular simulation is an extremely powerful tool that complements theoretical research and experiments. In order to be able to apply molecular simulations routinely, it is imperative to have efficient software that is easy to use, and physical models that are reliable and well-characterized in terms of their predictive power. We are working on a new model where the predictive power is the main design criterion, and where we will provide quantitative analyses of the quality of the models. In addition we are involved in large scale application of existing models in the context of studies of DNA-protein interactions. Both the method development and applications have so far been very successful in terms of publications. Neither of these would have been possible without the aid of NAISS computer resources.