Down syndrome, the most common single cause of human birth defects, produces alterations in physical growth and mental retardation; its early detection is crucial. Children with Down syndrome generally have distinctive facial characteristics, which brings an opportunity for the computer-aided diagnosis of Down syndrome using photographs of patients. In this study, we propose a novel strategy based on machine learning techniques to detect Down syndrome automatically. A modified constrained local model is used to locate facial landmarks. Then geometric features and texture features based on local binary patterns are extracted around each landmark. Finally, Down syndrome is detected using a variety of classifiers. The best performance achieved 94.6% accuracy, 93.3% precision and 95.5% recall by using support vector machine with radial basis function kernel. The results indicate that our method could assist in Down syndrome screening effectively in a simple, non-invasive way.