Abstract: Gabor filter bank is applied to textile online defect detection and defect image segmentation.The parameter selection in many proposed methods in this area is evaluated firstly by defining a discriminability function computing on some synthesized defect images，and thereby an efficient parameter selection method is proposed in real time constraint.It is argued that the outputs of the real part of Gabor filters are the key factor for defect detection；the orientation angles of the filters are selected as horizontal and vertical direction，in which most defects appear；the radial center frequencies are selected according to the instinct frequency of the texture；and the length of the Gabor filters depends on the periodicity of the texture as well.Although the number of the Gabor filters decreases to four within the real time constraint，it is illustrated that most kinds of defects are correctly detected and segmented by the proposed filter bank.