Introduction to Life Cycle Assessment and Ecoinvent: Life Cycle Assessment (LCA) models the complex interaction between a product and the environment form cradle to grave. The aim of the LCA is to capture the environmental impact of a product and helps with decision making. One of the background databases for LCA is Ecoinvent. Studied Problem: In Ecoinvent, the data for the electricity generation mixes in a geographical area is generated by a log-normal distribution. The aim of this thesis is to provide an analysis of the uncertainty (a total range a value can take with a corresponding probability – probability density function) of the log-normal distribution in Ecoinvent and to promote a new approach for sampling data. The reason for studying the uncertainties for the electricity generation mixes is to give a better approximation how much electricity contributes to overall emissions in a LCA. The Methodology and Approach: The thesis covers two interlinked parts. The first part evaluates the fit of the log-normal distribution to real data. For this, the different histograms are compared, the ANOVA test is performed for a difference in the seasons, the hartigans dip test is performed, and to calculate the absolute percentage difference the Method of Moments (MoM) of the real data is compared to Ecoinvent. This is accompanied by the goodness of fit tests. Additionally, a linear correlation and cross correlation is applied. The linear correlation is applied in the second part and the cross correlation should provide as an additional evidence that the energy types are correlated. In the second part the correlations are applied in Brightway v 2, and we perform the correlated sampling. The results from correlated sampling, sampling from the parent data and the data in Ecoinvent v 2.2 and Ecoinvent v 3.2. are compared. (for a better overview consult chapter 9.2) Findings and Results Results of the statistical analysis showed that the log-normal is not a good fit for the energy mixes, since 35/56 distribution were at least bimodal. Additionally, the uncertainty for the different distribution mixes is underestimated. The standard deviation of Ecoinvent is on average 1100% off compared to the MoM. Correlated sampling was not possible, due to programming limitations and time. The sampling from the parent data showed that the uncertainty for the impact category of the global warming potential was underestimated. For the impact category ReCiPE and Ecotoxcicity, this was not the case, because they are more prone to other substances. Conclusions • The uncertainty in Ecoinvent is underestimated and could be improved at low cost. • A multimodal distribution fits better (in most cases) than the log-normal distribution. • There are some significant correlations between different fuel types. Recommendations • Improving the uncertainty in Ecoinvent by increasing the standard deviation. • Further research in correlated sampling for Ecoinvent. • Research in implementing a bimodal or multi-modal distribution into Ecoinvent. Limitations of this report • Active forecasting (modelling) was not investigated for Ecoinvent. • Correlation analysis did not consider partial correlation and multi-colinearity. • Implementing a standardized bimodal distribution (which is difficult) has not been covered.