Abstract. This paper presents a detailed empirical study of local search for Boolean satisfiability (SAT), highlighting several interesting properties, some of which were previously unknown or had only anecdotal evidence. Specifically, we study hard random 3-CNF formulas and provide surprisingly simple analytical fits for the optimal (static) noise level and the runtime at optimal noise, as a function of the clause-to-variable ratio. We also demonstrate, for the first time for local search, a power-law decay in the tail of the runtime distribution in the low noise regime. Finally, we discuss a Markov Chain model capturing this intriguing feature.Designing, understanding, and improving, local search methods for constraint reasoning, and in particular for Boolean satisfiability (SAT), has been the focus of hundreds of research papers since the 1990's and even earlier. For SAT, techniques such as greedy local search, tabu search, solution guided search, focused random walk, and reactive or adaptive search have led to much success. Specifically, Walksat [7] stands out as one of the initial solvers that introduced many of the key ideas in use today and, is still competitive with the state of the art.While many attempts have been made to understand the behavior of local search methods in terms of local minima, exploring "plateaus", the exploration vs. exploitation tradeoff, etc., our formal understanding is limited mostly to relatively simple variants of local search, such as a pure greedy search, a pure random walk, or a combination of the two. This is not surprising as the techniques employed by Walksat and other state-of-the-art local search solvers are too complex to allow a formal analysis in terms of, for example, a traditional Markov Chain. At the same time, there is a wealth of information available from observations of the behavior of local search methods on a variety of domains, most notably for random 3-SAT. There is either formal or anecdotal evidence of various features, such as Walksat scaling linearly at optimal noise but exponentially at sub-optimal noise, and suggestions that the runtime distribution of local search on a single random instance has an exponentially decaying tail. This work provides convincing empirical evidence in favor of, or even against, such