Recently, typical Global Navigation Satellite System (GNSS) receivers do not perform well in indoor scenarios. Thus, non-GNSS localization systems are required for these scenarios. Because the Wi-Fi access points (APs) are very common in most indoor environment, Wi-Fi becomes the most commonly used network infrastructure for indoor localization. For another, with the development of machine learning technique, extreme learning machine (ELM), as an emerging learning algorithm, was proposed with better generalization performance. So, in this chapter, the ELM based Wi-Fi indoor localization technique using received signal strength indicator (RSSI) is proposed. First, an ELM-based indoor localization algorithm by clustering analysis and two-stage feature extraction under semi-supervised conditions is proposed. In the off-line phase, the iterative self-organizing data analysis technique algorithm (ISODATA) is used to reveal the inherent nature of the RSSI measurements and obtains the class of each RSSI measurement. Then, the multi-kernel ELM (MK-ELM) method is proposed for classification learning and obtains the RSSI measurement classification function. For each data subset, a two-stage feature extraction algorithm is proposed for RSSI feature extraction. The kernel principal component analysis is used to obtain coarse feature. And then the deep learning network and ELM method are introduced to get refined feature for semi-supervised regression learning. At last, the position regression functions of each data subset are obtained. In the on-line phase, after the RSSI measurement classification and feature extraction of the received RSSI measurement, the position can be estimated with the corresponding position regression function. Second, an ELM-based multi-floor indoor localization technique is proposed. To maximize efficiency, a data preprocessing algorithm has been developed, aiming to efficiently extract out only the essential information from the vast amount of datasets and reduce the data dimension, and transform the floor-level datasets and positioning datasets of each floor into a proper structure that is suitable for the proposed ensemble ELM technique. The proposed solution is unique in that its off-line phase exploits multiple individual ELMs for all floors to generate a set of floor-level classification functions with the preprocessed training datasets, and for each floor, it exploits multiple ELMs for the data clusters to generate a set of position regression functions. The on-line phase executes a coarse localization step to estimate the floor by using the floor-level classification functions and a refined step to estimate the position on the floor by using the position regression functions.