Department of Mathematics, University of Coimbra, Portugal Reference databases for the accuracy assessment of land cover maps are assumed to represent the true land cover in a set of selected locations. Although, the elaboration of reference databases through field visits or photo interpretation of aerial images is a hard and difficult task due to uncertainty in the assignment of the most correct land cover class at each sample location because of landscape fragmentation. To capture the diversity of land cover in each pixel of the reference data, a linguistic scale composed by five linguistic values is used in this paper and a methodology to translate the linguistic values to fuzzy intervals is proposed, based on the responses of the photo interpreters to a set of chosen pixels. These fuzzy intervals are then used to build a fuzzy confusion matrix and fuzzy thematic accuracy measures are derived. The main goal of this study is to quantify the differences of perception of photo-interpreters when assigning a linguistic scale to the sample pixels of the reference database. The proposed approach is tested on a case study where the accuracy assessment of a map of Continental Portugal with five land cover classes, derived from MERIS images, is made.