This paper presents the 3-space model of conceptual blending that estimates the figurative similarity between Input spaces 1 and 2 using both their analogical similarity and the interconnecting Generic Space. We describe how our Dr Inventor model is being evaluated as a model of lexically based figurative similarity. We describe distinct but related evaluation tasks focused on 1) identifying novel and quality analogies between computer graphics publications 2) evaluation of machine generated translations of text documents 3) evaluation of documents in a plagiarism corpus. Our results show that Dr Inventor is capable of generating novel comparisons between publications but also appears to be a useful tool for evaluating machine translation systems and for detecting and assessing the level of plagiarism between documents. We also outline another more recent evaluation, using a corpus of patent applications.