Portuguese Hate Speech Twitter Dataset is a dataset of Twitter messages manually annotated for Hate Speech using a hierarchical structure of classes. 5,668 messages were collected on Twitter, from 1,156 distinct users and classified as containing hate speech using a hierarchical structure of classes. A multiclass and multilabel approach was considered. Two different formats of the dataset are provided, plus the hierarchy of classes. The text of the tweets is omitted in this dataset due to the conditions and terms of the Twitter API.