An image forming apparatus which can quickly remove remaining toner even after it has been emergently stopped when it forms an image by the use of an intermediate transfer member. An image forming apparatus for primary-transferring a toner image formed on a photosensitive drum to an intermediate transfer belt by a primary transfer roller, and secondary-transferring the primary-transferred image to a recording material to thereby form an image has a secondary cleaning portion for applying a bias to a plurality of fur brushes, wherein during ordinary image forming, biases of different polarities are applied to the plurality of fur brushes to thereby effect cleaning, and during a returning operation after emergent stoppage, biases of the same polarity are applied to the plurality of fur brushes to thereby effect cleaning.