This paper reexamines the event study methodology in finance. We consider a formal specification of an event study in terms of a system of abnormal returns and, in particular, emphasise the possible limitations of using a methodology when misspecification may be present. In the first section of the paper, the theory of the event study is reviewed, with reference to the definition problems associated with the measurement of abnormal returns, the conditional information set embedded in return expectations, the determination of the event window, and the similarity across events. A major insight of our paper is to emphasise the importance of conditionality, learning and convergence in the theory of event studies and in the evolution of abnormal returns. In one section of the paper, we look at the question of recursive residuals, residuals formed when the information set is updated period by period, and implicitly question why they have not been used in event studies. In another section, the focus is on the specification of learning models, both across events and within the event period. Furthermore, we provide some conjectures about how to measure the information in events. Finally, we bring together the issues into a coherent methodology.