We discuss an probabilistic error bound on the popular LASSO method, a popular convex optimization method that quickly chooses a fixed number of the most important variables, say S, from among a large number of variables, say P. An alternative solution is to test the all subsets of size S among the P variables but this direct solution is often computationally intractable. The bound shows that with high probability and under mild assumptions, LASSO's performance is of order log P away from the best possible solution's performance. In other words, LASSO often chooses a reasonably performing set of S variables. We show how an extension of this error bound is useful in choosing a subset of microphones to monitor a fixed target in a rectangular room.