Metropolis-Hastings (aka Markov chain Monte Carlo). Without this very simple (in principle) algorithm and its many descendants, much of modern applied Bayesian statistics would be impossible.