The sensory characteristics are evaluated by descriptive methods of sensory evaluation, such as the conventional profile. The sensory results from profiling present a large variability, due to differences between assessors on the one hand, and within-batch variability, that is differences between units of a sample, on the other hand. To date, differences between assessors are taken into account and sensory methodologies have been adapted for that. Nevertheless, within-batch variability has not been tackled, despite the challenge it represents. With the apple as a model, this PhD work addresses the issue of within-batch variability from two points of view: the measuring of variability and the reduction of variability, with the aim of improving the reliability of the sensory measure. First, the capacity of the panel to discriminate between apples from different cultivars and between apples, identified as different by an instrumental measure, within each cultivar has been observed. Second, in order to take within-batch variability into account, the standard models for analysis of variance used in sensory evaluation has been adapted by adding a fruit factor nested within the cultivar factor. This adaptation requires the sharing of each piece of fruit by several assessors. The application of this model has demonstrated the importance of within-batch variability and the consequence of model choice to obtain more relevant results. In addition, the determined model for analysis of variance has been adapted to distinguish between assessors' disagreement and scaling differences. Finally, a reduction of variability has been obtained through the sharing of fruits and after the homogenization by instrumental measurement, the latter depending on the cultivar and on the chosen measure. The results of the present work provide methodological keys to obtain reliable sensory results in the frame of the evaluation of products subject to biological variation.
