Utility companies consistently suffer from the harassing of Non-Technical Loss (NTL) frauds globally. In the traditional power grid, electricity theft is the main form of NTL frauds. In Smart Grid, smart meter thwarts electricity theft in some ways but cause more problems, e.g., intrusions, hacking, and malicious manipulation. Various detectors have been proposed to detect NTL frauds including physical methods, intrusion-detection based methods, profile-based methods, statistic methods, and comparison-based methods. However, these methods either rely on user behavior analysis which requires a large amount of detailed energy consumption data causing privacy concerns or need a lot of extra devices which are expensive. Or they have some other problems. In this dissertation, we thoroughly study NTL frauds in Smart Grid. We thoroughly survey the existing solutions and divided them into five categories. After studying the problems of the existing solutions, We propose three novel detectors to detect NTL frauds in Smart Grid which can address the problems of all the existing solutions. These detectors model an adversary's behavior and detect NTL frauds based on several numerical analysis methods which are lightweight and non-traditional. The first detector is named NTL Fraud Detection (NFD) which is based on Lagrange polynomial. NFD can detect a single tampered meter as well as multiple tampered meters in a group. The second detector is based on Recursive Least Square (RLS), which is named Fast NTL Fraud Detection (FNFD). FNFD is proposed to improve the detection speed of NFD. Colluded NTL Fraud Detection (CNFD) is the third detector that we propose to detect colluded NTL frauds. We have also studied the parameter selection and performance of these detectors.