The task of selecting suitable fonts for a given text is non-trivial, as tens of thousands of fonts are available, and the choice of font has been shown to be able to affect the perception of the text. Aiming to support the development of font recommendation tools, in this talk, we predict semantic font attribute vectors using k-NN regression with neural network embeddings as our similarity measure. Moreover, we create a typographical lexicon using affective associations of the words and fonts. The cross validation results and user evaluations show strong support for both the font vectors and the lexicon.