We propose an automatic system to recognize sign language using principal component analysis (PCA) and one-vs.-all support vector machines (SVM) classification. The algorithm was trained and tested using a total of 500 images of the five vowels. The method includes color information, to detect skin regions, hand segmentation, using morphological operations and filters, feature extraction in hand regions using PCA, and classification using SVM. A graphical user interface was implemented for real-time recognition. For this first approach, the system was optimized for working with the five vowels showing results of a testing accuracy above 80% and an execution time of 59 milliseconds per frame.