There are random and irregular and variable properties in power quality disturbance detection signal that make it more difficult to identify the signal. This paper improved the original form of morphological un-decimated wavelet decomposition (MUDW) and applied it to power quality detection to locate the disturbance signals. The improved MUDW scheme contains broad open-close or close-open filter and morphological gradient which detect mutations on the top and bottom edges of signal. The scheme meets the signal reconstruction conditions. This scheme used MATLB to detect transient or steady state with single or composite disturbance signals and compared with the existing form of sampling wavelet. The result shows that the new MUDW scheme can recognize and detect the signal and has good anti-noise performance.