The study adopts Kansei Engineering and back-propagation neural network to forecast the student groups' Kansei evaluation regarding form design of perfume bottles, and the results demonstrates that the cognition are inconsistent between the students with a major in design and the others, with the only exception of the word pair of “Stylish-Classic”. The learning modules of back-propagation model must be inspected under the consistence of cognition. Consequently, the resulting accuracy rate merely reaches 20% due to the diverse cognition. For a better prediction performance, a greater amount of training and learning processes for various groups are required for the back-propagation model.