PURPOSE: To predict defect generation after the product is delivered by inputting manufacture inspection items as to products of the same kind and the numbers of defects of the inspection items to a neural network.

CONSTITUTION: Data on the manufacture inspection items in manufacture and the numbers of defects of the inspection items of an inspection line and actual result data on the time up to the generation of the defects after delivery are gathered from a person in charge of product service after a manufacture line, the inspection line, and the delivery and stored 1 in a storage file, and a verifying process 3 is performed as to defect data on an item which is selected 2. Then a normalizing process 4 is carried out, a learning pattern that the neural network learns, is generated 5, and learnt data are stored 6 in a learnt data file 6. Then, current data are inputted from the manufacture line and inspection line and the neural network which has learnt performs a predicting process 7 for defect generation after delivery; and a reverse normalizing process 8 is performed for the prediction result, which are put back to units of output data, so that the predicted value is outputted 9.