Classical methods of image processing were combined with neural network techniques in order to determine relationships between human judgements on the quality of pot plants and physical measurements. First, 2 images of a pot plant are taken in different positions ; the images are segmented into regions for flowers, leaves and background. Several characteristics are determined for each of the regions of the flowers and leaves. Statistical methods, such as principal component analyses and stepwise selection of variables by multiple linear regression, are used to condense information. The results are the inputs for the multilayer neural network. The performance yardstick for the model is the Pearson correlation with the mean judgement. It is compared with the individual experts to determine whether it is possible to replace an expert by the model.