Symbolic data analysis (SDA) is an underdeveloped statistical method in which the individual ‘data points’ for analysis are themselves distributions. How to construct and then analyse these ‘symbols’ is an ongoing research problem. This talk will briefly introduce the ideas behind SDA and propose a new general modelling approach for this type of data. The flexibility of our method is discussed and illustrated for interval and histogram-valued symbols. It will then be demonstrated how it can be used to fit extreme value distributions with an arbitrarily large numbers of classical observations. This procedure can be shown to offer one way to fit max-stable process models using pairwise (or higher) composite-likelihoods with an arbitrarily large number of spatial locations, which would otherwise be computationally prohibitive.