The application of sentiment analysis in social networks supports the understanding of complaints and claims of users' comments. To train the models that automate this analysis, it is important to construct guidelines that generate a more robust corpus. As far as we know, no related work of guidelines for spanish comments annotation has been found. We propose a method to construct guidelines to annotators reach a consensus in the entire annotation process of spanish comments from social networks. We annotated 3259 spanish comments using our guidelines, where the concordance analysis from our annotators was 84%. We employed our corpus and eight baseline classifiers for sentiment analysis detection, achieving 78.63% as the highest F1-Score with Multilayer Perceptron. Our method is useful to tackle labeling spanish comments which can be used in NLP tasks such as sentiment analysis.