This paper describes two image classification tasks, carried out in the context of the interdisciplinary Digital Humanities project ChIA (Accessing and Analysing Cultural Image with New Technologies). On a set of selected still life food images from the Euro-peana collection, two classification rounds were carried out by five human annotators. On the one hand, concrete food objects were annotated; on the other hand, also abstract cultural features. The degree of precision in the description of labels was varied. In the first annotation round, no description of labels was provided and annotators relied solely on their interpretation or intuition. In the second annotation round, annotators were given concrete definitions of the labels. Besides, the annotators varied in age, gender and cul-tural background. The aim of these tasks was to determine whether patterns in the clas-sified data would emerge and if so, which parameters would be decisive. As part of the evaluation the inter-annotator agreement was calculated. Preliminary results suggest that the identification of cultural features in images is highly subjective. Agreements were higher for the classification of concrete objects than for abstract features. Some tentative patterns emerged with regard to gender, but a larger annotated data set is needed to draw further and more definite conclusions.