Loss data on hazard events can provide valuable information to both researchers and decision makers. Storm Data, the National Weather Service's (NWS) database, provides a comprehensive collection of loss data for all U.S. weather-related hazard events. This information is used by a range of policy makers, researchers, and the public for a variety of purposes. Yet there has been little analysis of the process used to generate Storm Data loss data, which leads to several questions about the reliability and validity of the information and the process used to create it. What sources are used for estimating damages? What types of damages are included in damage estimates? When and how are damage estimates adjusted? Is there sufficient time and training for those gathering and entering the data – i.e. NWS employees? How confident do NWS employees feel estimating damages? In order to answer some of these questions, we conducted a survey of all 123 NWS Weather Forecast Offices (WFO) through a two-part, independent (i.e., administered outside of the NWS system) and confidential online survey in order to elicit the most candid responses possible. Part A of the survey, implemented in July 2008, requested a collective response from each WFO on qualitative questions related to the loss estimation methodology at each WFO, as well as the causes of problematic loss data. Part B of the survey, implemented in September 2008, asked quantitative questions about the specific loss estimation process for randomly selected recent events. Part B was directed to the individual responsible for each specific event's loss estimate. In addition, Part B differentiated between events (a) where loss estimates greater than zero dollars were entered, (b) events where loss estimates of exactly zero dollars were entered, and (c) events where no information was entered, to gather insight into the reasons for these differentiations. Results from the study will be used to help the NWS Performance Branch consider efforts to improve the reliability and consistency of future damage data. In this paper, we present preliminary results from this survey, along with a discussion of our methodology and an overview of the Storm Data data generation process, in hopes of facilitating a discussion on the importance of reliable loss data with respect to its end uses.