This article, an efficient system for texture DEFECT detection based on curvelet transform is presented. The main idea is to model the DEFECTs in the texture image as one-dimensional discontinuities.Based on this idea, the curvelet transform is the most efficient method for describing DEFECTs. First, in the learning phase, training samples of intact and DEFECTed blocks of the texture image are collected and transformed to the curvelet domain. Next, for each block a feature vector based on curvelet sub-bands is extracted and using a proposed method some important and effective features are determined for the desired texture. Then, a proper threshold for detecting DEFECTed from intact blocks is determined. In the performance phase, a vector containing the important features from each block of the texture is extracted and then the block by is classified. The results of simulation show that the proposed system is superior to the mean shift method in detecting DEFECTed texture blocks, and is less sensitive to the type of texture.