The work aims to improve the accuracy for detection of defects in surfaces of steel billets. Because of the presence of the scale-covered on the billet surface, its characteristics such as brightness and texture in the background region were inconsistent. Moreover, the similarities in the gray-levels of the defect and defect-free regions made it very difficult to accurately detect the defects. In order to solve the above-mentioned problems and to detect surface defects more effectively, a method (based on Opt-Gabor) for detection of defects in the surfaces of steel billets by analyzing the characteristics of the metal billet surface defects to classify such part surface defects into two types. In order to select the best four Gabor filter parameters, two evaluation functions were designed to maximize the use of the energy difference between defect-free and defect regions. Moreover, the dual-threshold filtering method was used to reduce the measurement errors caused by noise and pseudo defects. Forty images with defects were tested with three kinds of filtering algorithms. The experiment showed that the proposed method had a detection accuracy of 92.50%, 92.50% and 95.50% for corner cracks, thin cracks and pseudo crack respectively. Opt-Gabor algorithm can defect the billet surface defects in a more accurate way according to the two kinds of different cracks classified. With respect to measurement accuracy, Opt-Gabor algorithm is slightly superior to other algorithms.