A problem-solving procedure in which candidate solutions are generated from each possible goal of a knowledge-based system and evidence is then gathered to determine whether each is acceptable for the current situation. See also backward chaining.