We describe an atomistic method for computing the viscosity of highly viscous liquids based on activated state kinetics. A basin-filling algorithm allowing the system to climb out of deep energy minima through a series of activation and relaxation is proposed and first benchmarked on the problem of adatom diffusion on a metal surface. It is then used to generate transition state pathway trajectories in the potential energy landscape of a binary Lennard-Jones system. Analysis of a sampled trajectory shows the system moves from one deep minimum to another by a process that involves high activation energy and the crossing of many local minima and saddle points. To use the trajectory data to compute the viscosity we derive a Markov Network model within the Green–Kubo formalism and show that it is capable of producing the temperature dependence in the low-viscosity regime described by molecular dynamics simulation, and in the high-viscosity regime (102–1012 Pa s) shown by experiments on fragile glass-forming liquids. We also derive a mean-field-like description involving a coarse-grained temperature-dependent activation barrier, and show it can account qualitatively for the fragile behavior. From the standpoint of molecular studies of transport phenomena this work provides access to long relaxation time processes beyond the reach of current molecular dynamics capabilities. In a companion paper we report a similar study of silica, a representative strong liquid. A comparison of the two systems gives insight into the fundamental difference between strong and fragile temperature variations.