A numerical procedure to allocate qualitative observations of soil individuals to soil classes is proposed. It takes into account both the fuzzy nature of soil classes and the imprecision of the cursory descriptions of soil individuals. The procedure is based on possibility theory and it uses an algorithm for fuzzy pattern matching. It includes three steps: (i) translating soil class and soil individual descriptions into a set of membership functions and possibility distributions of soil attributes, respectively; (ii) assessing the similarity between soil individuals and candidate soil classes by possibility and necessity measures; and (iii) defining a threshold of possibility and necessity to allocate individuals to candidate soil classes. This allocation procedure is tested with eight major soil classes and 83 soil individual descriptions extracted from the 1:250000 scale soil database of the Languedoc-Roussillon (south of France). The results show that a fairly good agreement with experts is possible although some allocations remain imprecise or impossible. Further improvements are suggested regarding both the procedure itself and the data used