A new method which is based on bidimensional empirical mode decomposition（BEMD） and wavelet thresholding was proposed for the noise removal in medical image of magnetic resonance image（MRI）. Namely the image was decomposed into the intrinsic mode function (IMF) domain. Then the wavelet thresholding was used to remove the noise in the IMF. After the characteristic of the wavelet hard thresholding and the wavelet soft thresholding was analyzed

an improved wavelet thresholding which overcomes the shortcoming of the custom wavelet thresholding for denoising was introduced. In addition to remove the noise in MRI image

the experimental results show that our method had preserved the details of MRI image. It was propitious to medical diagnoses.