The "max_df" parameter of TfidfVectorizer in scikit-learn is used to set an upper limit on the term frequency within a document. Terms that occur more frequently than the specified value are ignored during the vectorization process.