Classification　of　imagery　motor　tasks　is　the　main　challenge　of　analysing　electroencephalography　(EEG)　data　from　a　brain-computer　interfaces　(BCI)　system.　The　noise　and　artifacts　recorded　by　BCI　system　frequently　corrupt　imagery　motor　EEG　data　and　reduce　classification　accuracy.　In　this　paper,　wavelet　denoising　algorithm　is　proposed　to　reduce　noise　from　motor　imagery　EEG　data　and　a　power　spectral　density　(PSD)　feature　selection　method　is　used　to　improve　classification　accuracy.　Experimental　results　show　that　the　classification　accuracy　of　the　proposed　method　is　significantly　improved　compared　to　the　same　PSD　feature　selection　method　without　wavelet　denoising.　This　result　also　certainly　indicated　that　wavelet　denoising　algorithm　successfully　purified　motor　imagery　EEG　data　and　made　classifying　features　more　prominent.