An electrostatic latent image is formed on a rotary image carrier. A developing roller is separatably abutted on the image carrier to supply toner onto the image carrier to make the latent image visible as a toner image. An intermediate transfer member is adapted to temporarily hold the toner image. A first transferer presses the intermediate transfer member against the image carrier to define a primary transfer position therebetween, so that the toner image on the image carrier is transferred to the intermediate transfer member. A second transferer is separatably abutted on the intermediate transfer member to transfer the toner image on the intermediate transfer member to a recording medium. An operation for forming the latent image is started after a predetermined time period elapses since a toner attached on at least one of a first region of the intermediate transfer member, on which a toner image to be transferred onto the recording medium is not transferred, and a second region on the image carrier corresponding to the first region has passed through the primary transfer position.