In this paper we use multidimensional scaling to analyse a corpus of texts and lexical items written in Linear A, an undeciphered script. We argue that more such work is necessary before Linear A can be deciphered (if this is indeed a possibility in the first place).