Easy access to social media enables users to express their opinions and ideology about various topics like news, videos, and personalities freely, without any fear, and often in an offensive manner. It is a vital task to detect comments with offensive language on social media platforms and relies on a complete and comprehensive tagged dataset. Therefore, in this paper, we introduce and make publicly available PerBOLD, a new Persian comment dataset collected from Instagram as a popular platform among Iranian. We follow a two-level manual annotation process in order to determine whether a comment has offensive language or not and fine-grained tags of different types of offensive language. Furthermore, we present some interesting aspects of data and analysis them.