Abstract. Most complete answer set solvers are based on DPLL. One of the constraint propagation methods is the so-called lookahead, which has been somewhat controversial, due to its high overhead. In this paper, we show characterizations of the problems for which lookahead is ineffective, and demonstrate, experimentally, that for problems that lie in the phase transition regions, search efficiency can be improved significantly by lookahead. This understanding leads to the proposal of a mechanism called adaptive lookahead, which decides when lookahead should be invoked dynamically upon learned information. This new mechanism is shown promising by the experiments.