This paper explores the feasibility of modelling concept concreteness perceived by humans and representing it in computational semantic lexicons, addressing an issue at the crossroads of computational linguistics, lexicography, and psycholinguistics. The inherent distinction between concrete words and abstract words in psychology has relied mostly on subjective human ratings. This practice is hardly scalable and does not consider the effect of polysemy. In view of this, we attempt to obtain a measure of concreteness from dictionary definitions comparable to human judgement, capitalising on conventional lexicographic assumptions and the regularities exhibited in the surface structures of sense definitions. The structural pattern of a definition is analysed and scored on a 7-point scale of concreteness ratings. The definition scores turned out to be quite effective for a dichotomous distinction between concrete and abstract concepts and more consistent with human ratings for the former. Beyond the two-way distinction, however, the results were more variable. The study has thus revealed the potentials and limitations of our approach, suggesting that different defining styles probably reflect the describability of concepts, and describability alone may not be sufficient for differentiating the degree of concreteness. The range of definition patterns has to be reconsidered, in combination with other inseparable factors constituting our perception of concreteness, for better modelling on a finer scale of concreteness distinction to enrich semantic lexicons for natural language processing.