This paper presents a novel deep learning-based indoor fingerprinting system, termed DeepFi, which uses Channel State Information (CSI) to improve the accuracy of indoor localization. The system architecture includes an offline training phase and an online localization phase. In the offline phase, deep learning is used to train a deep network, and a greedy learning algorithm is employed to reduce computational complexity. In the online phase, a probabilistic method based on radial basis functions is used to estimate the location. Experimental results show that DeepFi outperforms existing methods in two representative indoor environments, demonstrating its effectiveness in reducing location errors. The paper also discusses the impact of different parameters and propagation environments on the performance of DeepFi.This paper presents a novel deep learning-based indoor fingerprinting system, termed DeepFi, which uses Channel State Information (CSI) to improve the accuracy of indoor localization. The system architecture includes an offline training phase and an online localization phase. In the offline phase, deep learning is used to train a deep network, and a greedy learning algorithm is employed to reduce computational complexity. In the online phase, a probabilistic method based on radial basis functions is used to estimate the location. Experimental results show that DeepFi outperforms existing methods in two representative indoor environments, demonstrating its effectiveness in reducing location errors. The paper also discusses the impact of different parameters and propagation environments on the performance of DeepFi.