With the development of Markov Chain Monte Carlo (MCMC) methods it is now possible to perform accurate Bayesian inferences with extremely complicated models for data. If a simpler model is sufficient, then an efficient rejection sampling algorithm may exist, and would be preferable to MCMC. I will describe a procedure for simulating independent draws from the joint posterior distribution of the parameters of a 2-level Normal hierarchical model. Candidate draws for the unknown level-1 and level-2 variances are subjected to rejection, and conditional on accepted draws, the level-1 and level-2 mean parameters may be simulated directly. I demonstrate the algorithm by projecting season rushing totals for NFL running backs based on mid-season statistics.