The purpose of this research is to demonstrate a methodology for quantifying the variability and uncertainty in emission factors and emission inventories. Emission inventories are used for various policy-making purposes, such as characterization of temporal emission trends, emissions budgeting for regulatory and compliance purposes, and the prediction of ambient pollutant concentration using air quality models. Failure to account for variability and uncertainty in emission inventories may lead to erroneous conclusions regarding source apportionment, compliance with emission standards, emission trends, and the impact of emissions on air quality. Variability is the heterogeneity of values of a quantity with respect to time, space, or across a population while uncertainty arises due to lack of knowledge about the true value of a quantity. The sources of variability and uncertainty are distinct and hence variability and uncertainty affect policy- making in different ways. For example, variability in emissions arises from differences in operating conditions among different power plants. Uncertainty arises due to measurement errors, systematic errors, and random sampling errors. It is possible to reduce uncertainty by taking more accurate and precise measurements (i.e. reducing measurement error) or by taking a larger number of measurements (i.e. random sampling error). However, it is not possible to reduce variability. Therefore, in this research variability and uncertainty are treated separately. A methodology for simultaneous characterization of variability and uncertainty in emission and activity factors and their propagation through an emission inventory model is described. Variability was characterized using probability distributions developed on the basis of data analysis. The uncertainty due to random sampling error was characterized using parametric bootstrap simulation. A methodology for the quantification of variability and uncertainty in censored data sets containing below detection limit values was developed. This methodology is demonstrated for three case studies. In Case Study 1, the variability and uncertainty in the activity and emission factors for NO x emissions from selected coal-fired power plant systems was quantified based on data obtained from the U.S. Environmental Protection Agency. An illustrative partial probabilistic NO x emission inventory was developed for the state of North Carolina. In Case Study 2, the variability and uncertainty in the total short-term average emissions and in annual emissions of nine hazardous air pollutants (HAP) from a power plant was quantified by propagating the probability distributions for coal concentrations, boiler partitioning factors, and fabric filter partitioning factors through an emissions model. In Case Study 3, the effect of various levels of censoring on the variability and uncertainty in CO and HC emission factor data sets for diesel transit buses was studied. The main findings regarding the methodology demonstrated in this research include: (1) uncertainty due to random sampling error is substantial and in many cases was found to be of the same order of magnitude as the variability in the data set; and (2) the methodology developed for quantifying the variability and uncertainty in censored data sets is reasonably robust and accurate. The main insights obtained from the application of the methodology include: (1) the uncertainty in the total NO x emissions from selected power plants in North Carolina is ± 25 percent around the nominal value; (2) the uncertainty in the short-term average emissions of all HAPs from a power plant is substantially high in the upper percentiles (e.g., the width of the 95 percent confidence interval on the 95th percentile is 385 lb) than in the lower percentiles (e.g., the width of the 95 percent confidence interval on the median value is 60 lb) ; (3) the range of uncertainty in the annual average emissions is much wider than the range of variability in annual average emissions from one year to another; and (4) the uncertainty in the median value of censored CO and HC emission factor data sets increases as the level of censoring increases.