In this paper a new technique for defect detection in gray-level textured images is proposed. The first step of the algorithm is devoted to compute the local homogeneity of each pixel to construct a new homogeneity image denoted as (H-image). The second step consists in dividing the H-image into squared blocks and applying the discrete cosine transform (DCT) and then representative energy features of each DCT block are extracted. The defect blocks can be determined by a multivariate statistical method. Finally, a simple thresholding method is applied to extract defective areas. Simulations on different textured images and different defect aspects show good promising results.