The "ngram_range" parameter of TfidfVectorizer in scikit-learn is used to specify the range of n-grams (contiguous sequences of n words) considered during the vectorization process. It defines the lower and upper bounds for the n-gram sizes that will be included in the feature representation.