With the developing technology, smart systems have started to take place in our daily lives. Accordingly, it is very important for the systems that will actively participate in social life to adapt to social life properly. One of the most important steps of adapting to social life is communication. Facial expressions are one of the most important parts of communication that usually supports verbal communication. For this reason, many studies have been carried out on identifying facial expressions. The vast majority of these studies were carried out using datasets containing only adult faces. Conducting studies that do not involve the elderly and children may lead to the creation and development of highly biased smart systems. Therefore, this article focuses on detecting children's facial expressions. In order to detect facial expressions in children, a data set was prepared with images collected from search engines using keywords. By using the transfer learning method, the success of VGG16, ResNet50, DenseNet121, InceptionV3, InceptionResNetV2 and Xception models were evaluated and compared on this prepared data set. According to the evaluation results, the best result was obtained with the InceptionV3 model with an accuracy rate of 76.3% and an F1 score of 0.76.