Identifying subjective sentences and classifying the polarity of subjective sentences are two important tasks in sentiment analysis. Besides being a hot topic, there is still a lack of resources to perform the mentioned sentiment analysis tasks in the Portuguese language, with its syntactic specificities. This paper describes the identified challenges and next steps in an initial study regarding the classification of subjectivity and polarity of sentences with a small set of syntactic features extracted directly from the text. Our approach reached satisfying results in experiments with two classic machine learning models in four datasets consisting of user reviews from different domains.