Disclosed herein is a practical knowledge acquisition system, in which, when a diagnosis is made using a knowledge base in the form of an associative network turns out to be erroneous, the network is analyzed backwards from the node representing a desired solution, thereby detecting a node where the backward analysis route joins that analysis route in the network which has resulted in the erroneous diagnosis. The node where the two analysis routes join is identified as one possibility where an error has occurred in the diagnosis. The diagnostic knowledge applied at this node is seen as an candidate to be modified many modification plans for the incorrect knowledge is proposed, the diagnostic knowledge base can be modified with high efficiency.