Nowadays, indoor positioning is becoming one of the most important issues in smart cities. With the rapid progress of wireless communication and digital electronic technology, wireless sensor networks (WSNs) have been developed and are playing an important role in indoor positioning systems. The received signal strength indicator (RSSI) is adopted by most range-based localization algorithms. However, the positioning system based on the RSSI is vulnerable to environmental interference and the RSS itself is unstable. To tackle this problem, we propose an improved indoor localization based on the RSSI and general regression neural network (GRNN). In the raw data processing module, an improved average filter is proposed to make the raw data stable and reliable. Then, an improved weighted centroid localization algorithm (IWCLA) is proposed to revise the positioning result on the basis of maximum likelihood estimation (MLE). In the view of the complex and changeable indoor environment, an improved GRNN localization algorithm is proposed to achieve better applicability and higher positioning accuracy. The effectiveness of the proposed methods is verified in different cases through simulation and experiment studies.