The Projection algorithm is a heuristic algorithm that does not guarantee that the unknown motif will be found every time. We can increase the chance of success by performing a large number of independent trials to generate multiple guesses. In each trial, Projection makes a preselection of sets of length-l patterns called l-mers which are likely to be a collection of motif instances (filtering step) and refines them by some local searching techniques, e.g. EM algorithm, Gibbs Sampling etc (refinement step).