The leading eigenvalues and corresponding eigenfunctions of the transfer operator are essential for the analysis of Molecular Dynamics simulations. As the transfer operator is a self-adjoint linear map, they satisfy a variational principle and can be approximated by the method of linear variation similar to the one used in quantum chemistry. In this talk, I will explain this approach and present some first applications to small peptide systems. I will also explain that Markov State Models are a special case for a specific choice of basis functions, but that more general basis sets may potentially provide better results. Finally, I will discuss what needs to be done in the future to make use of the method when the system size grows.