Abstract: In this paper, an adaptive and multidirectional method for fabric defect detection based on bidimensional EMD was proposed. Delaunay triangular partition, radial basis function interpolation and bidimensional cubic spline interpolation were used and combined to implement the algorithm of bidimensional EMD, which was applied to decompose digital images of fabric into a set of sub-images, then sub-images contained defect features were selected to reconstruct a new image, and finally the defects could be distinguished from the binaryzation of the reconstructed image. Several different defective fabric images captured by an industrial line scan camera were analyzed using the proposed approach and the results demonstrated that features of defect contained in reconstructed images were distinguishing and had strong contrast against the background; defects could be recognized and located from binary results without other processes such as noise eliminating. This detecting method could be qualified to fabric defects.