This paper examines how monitoring the access and use of web-based museum collections can improve the overall quality of artifact data. All databases have quality problems that arise both from input errors and also from errors that emerge over time; museum databases are no exception. Indeed, the complexity and variety of the information recorded in museum databases actually adds to the probability of such errors occurring. Given the expense of checking through each record for errors, we are exploring alternative approaches that can contribute to the improvement of data quality for web-accessible museum database systems. Since 1997, the Spurlock Museum at the University of Illinois has been performing a complete re-inventory of its collection, developing in the process an extensive database of artifact data available online. This project has required the cooperation of museum staff members, various external experts, and dozens of part-time undergraduate student employees. During the process of data gathering, entry, and analysis, an unexpected phenomenon was observed: in accessing and using the data, individuals have consistently detected errors, reported them, and even volunteered corrections. However, these individuals were not directly involved in the task of searching for errors; they were instead working on other tasks. For example, a curator working remotely over the Internet to plan a new exhibit for the museum retrieved several artifact records using his web browser. In one of these records, he found an erroneous entry and sent an email to the museum's registrar, noting the error and suggesting a correction. The relative frequency of this kind of activity raises a question: can web-accessible museum collections incorporate mechanisms that will allow people browsing the data to report errors they detect and, should they wish, volunteer corrections? Although clearly this will not happen for every database record, we need to understand more about the process of error recognition and correction in order to facilitate it when it does occur. This paper will present the results of our study of unexpected error detection and correction in the Spurlock Museum artifact database. Through an analysis of this phenomenon, we explore how to encourage such data correction behavior explicitly, effectively and efficiently. We also note the importance of establishing and encouraging a community that will support collaborative data correction activities.