The accuracy of detecting the orange skin surface defect in lacquered furniture elements was tested. Textural features and an SVM classifier were used. Features were selected from a set of 50 features with the bottom-up feature selection strategy driven by the Fisher measure. The features selected were the Kolmogorow-Smirnow-based features, some of the Hilbert curve-based features, some of the maximum subregions features and also some of the thresholding-based features. The Otsu thresholding and percolation-based features were all rejected. The images of size $$300\,\times \,300$$300×300 pixels cut from the original, larger images were treated as objects. There were three quality classes: very good, good and bad. In the cross-validation process where the testing sets consisted of 90 and the training sets of 910 objects the accuracies ranged from 90% to 98% and the average accuracy was 94%. The tests revealed that more research should be done on the choice of features for this problem.