In this paper, we propose a new tire defect detection algorithm based on dictionary representation. The dictionary learned from normal images is efficient to represent defect-free images while it has low efficiency to represent defect images due to its capability of capturing key information. Unlike the conventional iterative solution with complicated calculation, the representation coefficients are obtained by multiplying the pseudo-inverse matrix of the learned dictionary and image patch. Moreover, the distribution of representation coefficients is very different between defect-free image and defect image. Therefore, the distribution difference of representation coefficients can be used as a discrimination criterion to detect the defective region. Experimental results demonstrate that the proposed method can accurately detect defects.