The fast-spreading use of web surveys in social surveying accentuates the methodological cornerstones related to sampling and user identification. While the obvious advantages of web survey methodology –cost reduction, automation of data collection, accuracy, validation, timeliness –in first hand are of a practical nature, the problems are of greater consequence to more basic issues of survey research, such as sampling, representative samples and respondent identification. The non-probabilistic convenience sampling of many web surveys traditionally poses serious problems to the power of generalization and prediction of the survey, but do these fundamental problems also indicate actual problems in concrete library user surveys. This paper presents the results of a study comparing various traditional sampling methods to web sampling. The study was carried out at The Regional Library of Northern Jutland in Aalborg, Denmark, during spring 2005. Three samples of library users were created as a web sample drawn by self-selection from the library web site, a convenience sample drawn among visiting users over two weeks, and a random sample drawn from the library registration of users. Respondents from all three samples were presented to the same questionnaire, and the composition of the samples as well as the results and predictions were compared in order to determine and analyze the effect of sampling methods. During the web survey an automated collection of data on respondent behaviour was conducted in addition to the traditional data collection. The analysis of these data gives an insight into how respondents navigate through the questionnaire as well as detailed information on unit and item non-response. The data are also used to identity repetitious replies or “ballot stuffing”through identification of respondents. The paper presents an analysis of the problems connected to respondent identification and repetitious replies