We use our labeled data to train a classifier that predicts whether a stroke represents a visible descriptive line, or a hidden or guide line. Our motivation for those two classes is that visible descriptive lines are the ones most present in presentation drawings, and should as such be of special interest for sketch-filtering algorithms. We use a non-linear SVM with a Gaussian kernel function, which we train to predict a stroke label given a set of stroke features -- length, speed, time, pressure, mean curvature. We normalize the length, speed and pressure within each sketch by dividing their values by the 95%-percentile values over all strokes in that sketch. Finally, for a given training set of sketches, we randomly select 1000 strokes to form the training set for the classifier. The images below show the clussification results, when we trained a single classifier with all features, using a subset of objects and designers for testing, and all others for training. We used the Wobble surface, Bumps and Tubes as test objects, and Student 2, Student 5, Student 9, Professional 3 and Professional 5 as test designers.