As feminist activism gains visibility online, women increasingly face gender-based abuse in digital spaces, much of it originating from the ”manosphere”—a constellation of anti- feminist online communities such as incels. While prior research has examined the preva- lence of misogyny in online communities, less attention has been paid to how misogynist disproportionately targets women along intersecting social identities, including race, class, sexuality, etc. Intersectionality describes how overlapping systems of oppression interact to shape the lived experiences of people (Crenshaw, 1991). Specifically, the current study focuses on intersectional misogyny, defined as misogyny that invokes harassment related to one’s race, ability, class, or other identities. This study contributes to current literature in this area through two research questions: (1) How effectively can fine-tuned language models be used to detect intersectional harassment in misogynistic online forums? and (2) How do categories of misogynistic expression cluster within manosphere discourse, and in what ways do these clusters coalesce around particular intersectional identity markers? To answer these questions, I draw on a dataset of over 716k posts from one of the most active and unmoderated platforms in the manosphere: incel.is forum, focusing on those matched to physical and sexual violence lexicons. I manually annotated 7,000 posts based on whether it contains the presence intersectional targeting hostile language. For RQ1, a supervised HateBERT model was fine-tuned to detect intersectional harassment within the corpus. For RQ 2, an unsupervised topic modeling pipeline using BERTopic was used to inductively extract thematic clusters from the entire corpus. Next, the supervised HateBERT classifier was overlaid on these four macro thematic clusters to compute the proportion of posts containing intersectional harassment. This two-step pipeline enable to identify intersectionally harmful post and to analyze how such intersectional harassment cluster within manosphere contexts. The best-performing classifier achieved an F1 score of 0.73, with significant recall im- provements through class weighting, highlighting the potential of NLP models to detect nuanced intersectional hate speech. Regarding RQ2, results show that intersectional ha- rassment is not evenly distributed across four emerging macro themes, including Violence Fantasy, Fatalistic Worldview, Body-Based Devaluation, and Socioeconomic Complaint. Limitations of the study lie in the scope of human annotation and classifier’s deficit ability of detecting implicit harassment that lack overt identity markers. Future research should develop benchmark datasets for implicit intersectionality and expand annotation frame- works to include a broader range of misogynistic expressions. These steps would allow for more robust models capable of uncovering how intersectional hostility evolves across digital subcultures over time.