Traditionally, utilities have been conservative regarding the infrastructure upgrade. This makes electric load forecasts critical for guiding electric utilities’ operation, planning, and maintenance decision-making. For a long while, utilities have relied on the point load forecast that provides a single expected value about the future to guide their decision process. However, when the load forecasting is conducted for the middle or long term, things become more uncertain. For example, while the weather for the next week may be predictable, the forecast of it for the next month(s) or year(s) become unreliable. An unreliable weather forecast is less likely to help on the load forecasting practices. Furthermore, the modernization of the grid has brought many changes to the electric utility industry. Changes such as the increasing penetration of renewable energy, the emerging distribution generation, and the bi-directional communication between the supplier and the end-users have brought much more uncertainties for the utilities’ load forecasting practices. The single-valued forecast or point forecast that gives a deterministic forecast about the future load does not provide any information on such uncertainties. In contrast, a probabilistic forecast that estimates the respective probabilities for all the possible future outcomes of a random variable provides opinions on the uncertainties. Although probabilistic forecasting have been studied for decades and researchers have tried to apply those techniques for probabilistic load forecasting (PLF) for the past several years, there are still many challenging issues in the PLF field, such as lack of quantitative evaluations on the PLF methods, ad-hoc selection of input scenarios for PLF, and the lack of practical guides for PLF. This dissertation dissects the PLF problem into three key components including the input scenario simulation, the modeling techniques, and the residual analysis. From the input scenario simulation perspective, this dissertation first raises a critical yet never answered question about the lack of methodological foundation for practicing probabilistic load forecasting through input simulation. Such lack of methodological foundation typically results in ad-hoc, judgmental and indefensible choice during the scenario generation step. This dissertation then investigates a framework to evaluate the effectiveness of three different temperature scenario generation techniques, namely the fixed-date method, the shifted-date method, and the bootstrap method, from which an empirical rule-of-thumb is developed to guide the temperature scenario generation practice for PLF. The establishment of this evaluation framework helps to lay a solid methodological foundation for practicing probabilistic load forecasting through temperature scenario simulation. The proposed framework can also be extended to evaluate and guide the practices on generating other input scenarios. The modeling techniques will still rely on the representative ones developed for point load forecasting but the focus will be on how to convert point forecasting results to probabilistic ones. From the residual simulation perspective, studying residual series itself is not anything new in load forecasting and its utility applications. Back to 1970s, for example, researchers were using mean and standard deviation to characterize uncertainties around electric load forecasts for probabilistic load flow analysis. However, most papers in the literature that modeled load forecast residuals assumed normality for the residual distribution. Such normality assumption has rarely been verified through any formal statistical test. This dissertation conducts a comprehensive study regarding the normality assumption of the residuals. It not only studies the residuals from load forecasting as a whole but further considers the potential impact of multiple seasonality existing in electricity demand on the normality assumption of residuals. Moreover, it comprehensively studies whether simulating residuals with the normality assumption improves the probabilistic forecasts which has never been studied before. Two case studies are used in this dissertation: (1) the first and primary case study is based on the system total demand of North Carolina Electric Membership Cooperation (NCEMC). NCEMC serves 93 out of the 100 counties of North Carolina. The weather condition varies quite a bit within NCEMC’s service territory; (2) the second case study is anonymous data from the load forecasting track of Global Energy Forecasting Competition 2014 (GEFCom2014). The data is public available which allows others to reproduce the results presented in this dissertation. Although only two case studies are presented in this dissertation to demonstrate the implementation and comparisons of the different PLF techniques, the PLF techniques discussed in this dissertation have outperformed the ones developed by several other PLF groups worldwide.From research perspective, this dissertation raises and answers questions regarding the methodological foundation for practicing temperature scenario generation and residual simulation techniques for PLF. The study in this dissertation also points out directions for future research in the PLF field. For example, how to generate and evaluate other weather scenarios such as relative humidity. From practices perspective, the findings from this dissertation study offer multiple practical options for utilities’ probabilistic load forecasting practices. Some of the findings presented in this dissertation have been in production use by utilities since 2012 with outstanding performance.