This paper introduces a method combining a 2-dimensional discrete wavelet transform (2-DWT) with energy threshold method to deal with the data compression in power quality detection. The high frequency signals and noise signals are first decomposed into three different directions by making use of the 2-dimensional Daubechies wavelet transform to process matrix data with row convolution and line convolution. The energy of signals is then concentrated on a few wavelet coefficients. The energy thresholds are set with the helps of mean-energy correction coefficients

which guarantee the left energy after compression over 99%. Thus

the distortion ratio of reconstructed signals is small enough and the noise of disturbing signals could be adaptively eliminated. Finally

results of simulation with six kinds of disturbing signals indicate this method has a higher compression ratio and a better ability to eliminate noise compared with those of wavelet packet transform.