Down syndrome (DS) is the most common genetic condition in the world and its early detection during the pregnancy has significant role for further assessments. Early intervention for infants and children with DS can make a major difference in improving their quality of life. This paper proposes an intelligent method for detection of Down syndrome in fetuses using ultrasound images and convolutional neural networks (ConvNet). Ultrasonography is a non-invasive way to diagnose DS. The important features in ultrasonography are Nuchal translucency, nasal bone, fetal heart rate, FMF angle, etc. Unlike the traditional methods, the proposed method doesn’t need for any feature extraction. In the proposed method, ConvNet extract new effective feature automatically from ultrasound images. Also, particle swarm optimization (PSO) algorithm is used for optimal selection of ConvNet structure such as number of convolutional layers, number of pooling layers, learning rate and so on. The proposed method tested on Wisconsin University midwifery clinic and the obtained results showed that the proposed method has excellent performance in DS detection.