In a vehicle equipped with a torque converter (4) having a lock-up clutch (3), learning control for obtaining a learning value (L_n) on the basis of meet-point information, with which the lock-up clutch (3) initiates torque transmission, is carried out. After acquiring the current learning detection value (M_n), a meet-point learning control unit (12c) calculates the current detection error (E_n) on the basis of the difference between the current learning detection value (M_n) and the previous learning value (L_(n−1)) that is stored. When the current and previous detection errors (E_n) and (E_n−1) have the same plus/minus sign, the current learning value correction amount is set to a larger value if the absolute value |(E_n−1)| of the previous detection error (E_n−1) is large than if the absolute value |(E_n−1)| is small. The sum of the previous learning value (L_(n−1)) and the current learning value correction amount is set as the present learning value (L_n).