Disclosed is a technique for compensating for an abnormal output of a resolver. More specifically, a central processing unit (CPU) sets a current motor position angle before compensation θn,ORG as a current motor position angle θn and obtains a motor position change Δθn[rad] between a current sampling [n] and a previous sampling [n−1] and a motor position change Δθn-1[rad] between the previous sampling [n−1] and a more previous sampling [n−2]. Subsequently, a variable A is calculated based on the above angles. The CPU determines whether to perform the compensation by comparing the calculated variable A and a calibration variable K and calculates a current motor position angle for compensation θn[rad]. Finally, the CPU compensates for the absence of motor rotor position information with the calculated current motor position angle for compensation θn[rad].