In order to construct a novel design method for interior tiles, the re
lationship between image words (sensory evaluation) and total evaluati
on for interior tiles was studied. At first, the model for estimation
of the total evaluation from image words was constructed using a fuzzy
neural network (FNN) and MRA (multi regression analysis). The FNN mod
el could estimate the total evaluation more correctly than the MRA mod
el. Secondly, a FNN model for the estimation of the total evaluation f
rom analytical data for interior tiles was constructed. This model was
less accurate than the FNN model from image words, but could estimate
within 10% errors. Lastly, in order to improve accuracy of the estima
tion, a FNN model for estimation of the total evaluation from both ana
lytical data and image words was constructed. This model showed the hi
ghest accuracy among all models. ''Jyouhinna-gehinna'', ''shareta-yabo
na'' among image words and ''average value of green color'', ''arithme
tical mean deviation'', ''peak count'' among analytical data were need
ed to construct the model for estimation of the total evaluation.