Dynamic voltage restorers (DVRs) are used to protect sensitive loads from the effects of voltage sags on the distribution feeder. This paper presents and verifies a novel voltage sag detection technique for use in conjunction with the main control system of a DVR. In all cases it is necessary for the DVR control system to not only detect the start and end of a voltage sag but also to determine the sag depth and any associated phase shift. The DVR, which is placed in series with a sensitive load, must be able to respond quickly to a voltage sag if end users of sensitive equipment are to experience no voltage sags. A problem arises when fast evaluation of the sag depth and phase shift is required, as this information is normally embedded within the core of a main DVR control scheme and is not readily available to either users monitoring the state of the grid or parallel controllers. Previous research presented an additional controller, which required phase and sag depth information to manipulate the injection voltage vector returned by the main controller in order to prevent the DVR injection transformers from saturating. Typical standard information tracking or detection methods such as the Fourier transform or phase-locked loop (PLL) are too slow in returning this information, when either applied to the injection voltage vector, or to the supply voltages directly. As a result of this the voltage sag detection method in this paper proposes a new matrix method, which is able to compute the phase shift and voltage reduction of the supply voltage much quicker than the Fourier transform or a PLL. The paper also illustrates that the matrix method returns results that can be directly interpreted, whereas other methods such as the wavelet transform return results that can be difficult to interpret.