Latent images of different colors are superimposed on an intermediate transfer belt of an endless type from image carriers, to form a full color toner image. A tension roller maintains a tension of the intermediate transfer belt. A driving roller drives the intermediate transfer belt. A biasing unit applies the tension to the tension roller. A holding unit holds the tension roller. An image reading sensor detects the toner image formed on the intermediate transfer belt. The image reading sensor is disposed in the holding unit, in an opposite position to the tension roller with respect to the intermediate transfer belt.