In this paper we present a set of annotated data and the results of a number of unsupervised experiments for the analysis of sentiment in Latin poetry. More specifically, we describe a small gold standard made of eight poems by Horace, in which each sentence is labeled manually for the sentiment using a four-value classification (positive, negative, neutral and mixed). Then, we report on how this gold standard has been used to evaluate two automatic approaches for sentiment classification: one is lexicon-based and the other adopts a zero-shot transfer approach.1