A multipoint autofocus system in which a pair of object images, obtained by pupil-division of an object light bundle in each detection area, are projected onto different zones of a sensor and are converted into image signals, so that a defocus amount can be determined based on the image signals, the multipoint autofocus system including intersecting detection areas whose detection areas partly overlap and a plurality of sensor arrays of the sensor corresponding to respective the intersecting detection areas; and a selection device for selecting a defocus amount whose absolute value is smallest out of defocus amounts obtained by each sensor array of the intersecting detection areas.