The ubiquitous commercial Wi-Fi has brought unlimited possibilities to the smart city and the Internet of Things. Wi-Fi device-free sensing technology has received more and more attention in recent years. Counting the people in queuing is an uneasy task due to labile information. Most current counting schemes have existed in an ideal environment with idealistic people's behavior. In this article, we propose a more realistic counting scheme called Que-Fi, a queue number identification system based on Wi-Fi channel state information and a deep learning network. In the proposed Que-Fi scheme, the nonnegligible interference of human motion and the surrounding environment is first analyzed based on the Fresnel zone. Then, we proposed a static model based on the convolutional long short-term memory fully connected deep neural network in order to overcome the interference. A dynamic Que-Fi model is proposed to identify the entering/leaving people's behavior and update the counting number. In this article, different preprocessing methods are analyzed and compared to test and evaluate the proposed Que-Fi. Experiments have shown that the proposed Que-Fi outperforms the traditional support vector machine and provide accuracy up to 95% and 96.67% for static and dynamic models, respectively.