Abstract: In previous research on addition and subtraction word problems, people argued that part-whole knowledge plays a crucially important role, which is called part-whole hypothesis. However, there has been little convincing evidence to show the substantial influence of part-whole knowledge on children’s performance. The present experiment was to test the hypothesis. Children in the forth grade of primary school and undergraduates were asked to solve one-step addition word problems (correct answer is “a+b”, “a” referring to the first quantity occurred in problem text, “b” to the second quantity). If the part-whole knowledge that is a type of abstract logic-mathematical knowledge discarding the concrete features of object sets, mediate children’s thinking, it could be predicted that the correct answer was unanimously “a+b” or “b+a” without regard to the concrete features of object sets and the trend was more pronounced for undergraduates. The results showed that the concrete features of known object sets systematically influenced the surface of the correct answer, and especially the influence became more and more salient with age and practice, which was called hypothesis of problem familiarity on the surface of correct answer. The results were totally inconsistent with predictions. Therefore, it is necessary to doubt the part-whole hypothesis. The present research accounted for the results based on the position of object set in mental representation and representational strength. The instructional implication was discussed at the end of this paper

Key words: part-whole knowledge, addition word problems, arithmetic word problems, addition, subtraction, additive commutativity