In this paper, a similarity relation between two predicates is defined first. To a given predicate, the set of action for the predicate can be obtained by the similarity relation. Then, the domain knowledge is extracted from the common fluent in preconditions and effects of all actions for each set of action, and the formalism for the domain knowledge is given. Finally, the contradictions in the initial states and the goal states in a particular planning problem with domain knowledge can be discovered. The strategy of extracting the domain knowledge is integrated in the planner StepByStep, and the domain knowledge is the necessary theory when one predicate by action is realized in the planner.