Eversource Energy, formerly Public Service of New Hampshire (PSNH),has worked closely with Plymouth State University (PSU) in the past, and present, to better predict weather-related power outage events and maximize the efficiency with which Eversource responds to them. This research paired weather data from thirteen stations throughout New Hampshire, Vermont, and Massachusetts with Eversource Trouble Report and Unsatisfactory Performance of Equipment Report (TRUPER) data in an effort to quantify weather situations that lead to power outages. The ultimate goal involved developing a predictive model that uses weather data to forecast the magnitude of power outages. The study focused on the Eversource Western/Central service territory and utilized data from 2006-2010. The first four years, 2006-2009, were analyzed using Classification and Regression Tree (CART) statistical analysis. The results of this CART analysis trained a predictive model, while the fifth year, 2010, served as the testing set for the predictive model. To conduct the statistical analysis, a database was created pairing TRUPER reports with the closest available hourly weather observations. The database included nine weather variables matched with three variables from the TRUPER data: 1) customers, 2) customer minutes, and 3) outage duration. While the entire Eversource service territory saw 91,286 TRUPERs from 2006-2010, the Western/Central service territory, the focus of this study, accounted for 29,430. Before conducting the CART analysis, correlations between single weather variables and TRUPER data were calculated and, in general, proved weak. In addition to analyzing the complete four-year training data set, many portions/variations of the data set were analyzed. The analyses included a yearly analysis, time lag analysis, cold/warm-season analysis, and a single-station analysis. Although individual years and smaller data sets showed moderately higher correlations between weather and outage data, consistent relationships throughout the data set were fairly weak. CARTs were then created to examine the joint effect of the entire set of weather variables, such as interactions and nonlinear relationships, to improve the overall predictability of power outages. After creating the trees from the four-year training data set, their predictive ability was tested using the final year of data. The CART predictive models showed that among Eversource TRUPER variables, the hardest to predict was customers per TRUPER. The best performing model predicted customers per TRUPER to an average error of 96 customers, or a percent mean average error (PMAER) of131% of 2010 customers per TRUPER. This result could deal with the high variability seen in customer outages per TRUPER, across a single weather event, driven by widely-varying population and customer density. The most accurately predicted TRUPER variable, outage duration, saw average PMAER values of 60% of the mean (e.g., if mean duration per TRUPER for the year was 100 minutes, the model would miss on average by 60 minutes). Overall, the model results show surface weather data has a weak correlation to the TRUPER variables analyzed. The model can predict situations when one would expect longer duration outages but is unable to accurately predict the magnitude of these variables. When adapting the predictive models to smaller portions of the data set, warm-season data showed the greatest predictability, considerably outperforming the other data sets (cold-season, 2006-2009, and single station). Cold-season showed the greatest volatility and, not surprisingly, proved the most difficult to predict.