The literature survey on dividends shows that more explanation is needed though existing theories contribute considerably to the understanding of dividends including the rationale of paying dividends, smoothing of dividends, and greater frequency of dividend payments compared to stock repurchases. The purpose of this thesis is to explain the dividend phenomena using a reputation model in the world of imperfect information. In a two-period version of the model, the rationale of paying dividends and smoothing is analyzed. However, the reputation effect and the time pattern of dividends cannot be easily accommodated. We extend the model into a three-period model in which the behavior of dividends through time, and the market response on the change in dividends are analyzed. The reputation effect is incorporated such that the reliability of signalling improves with the passage of time. New empirical results which support the reputation model are presented. The market response to the initial dividend announcement is analyzed using all the cases of dividend initiation recorded on the CRSP daily stock files from July 2, 1962, to December 31, 1987. For firms which announce dividends during their first year on a major exchange, the abnormal return on the announcement date of initial dividends is not significant. Also, the change in the magnitude of the market response to the change in dividends through time is analyzed. We group the announcements of dividend increases according to the number of years passed after the initiation of dividends. The market's response to unexpected dividend increases in earlier periods is smaller than the case of unexpected dividend increases in later periods. The reputation model adds a new perspective to the analysis of dividends: the time aspect of dividends, that is, the effect of a consistent stream of dividend payments on the value of a firm. This model has direct applicability to all areas of dividends to better explain the dividend phenomena.