Much of the research on MIS implementation which has been conducted in the past decade has focused on identifying and measuring the organizational characteristics which appear to be particularly conducive to either success or failure of system development efforts. While such research is useful in providing insight about the implementation problem, it provides little guidance for the management of ongoing implementation efforts. The study described in this paper attempts to address the implementation management question by exploring the use of MIS users' pre-implementation expectations about a system as indicators of the likely success of that system. System development efforts can be viewed as multi-stage processes. During the first of the stages, Definition, most of the key decisions about the system as the user will see it are made, e.g., system goals, scope, overall approach. The Definition stage, however, typically accounts for no more than 25% of the resources required for system development. Thus, the decisions which will have the greatest effect on the users' acceptance or rejection of a system are made prior to the bulk of spending on the project, and an assessment of the project's probability of success or failure should be possible at that time. The results of a number of implementation studies suggest that implementation failure is more likely when users hold unrealistic expectations about a system. Research in other areas, especially product evaluation and job satisfaction, also shows a connection between realism of expectations and outcomes (e.g., satisfaction). Thus, user expectations held at the end of the Definition stage might serve as early warning indicators of MIS implementation outcomes. If these expectations prove to be reliable indicators of subsequent success or failure, it would enable system developers to diagnose likely problems and to take corrective action at an early project stage. This paper reports on a longitudinal study of user expectations as predictors of project success or failure. The results strongly suggest that users who hold realistic expectations prior to implementation are more satisfied with the system and use it more than users whose pre-implementation expectations are unrealistic. While the results are encouraging, further research is necessary in a number of areas—e.g., better definition of key expectations, simpler tools for measuring expectations, proper timing of expectations measurement—before reliable instruments for measuring expectations in ongoing projects will be available. The paper outlines, however, some steps which can be taken now to help assure that potential system users develop realistic expectations.