This paper empirically investigates the relationship between retirement duration and cognition amongst older Irish women using micro-data collected in the third wave of The Irish Longitudinal Study on Ageing. Ordinary least square (OLS) regression estimates indicate that the longer an individual has been retired, the lower their cognitive functioning, holding constant other factors thought to affect cognition (such as age, education and early- life socio-economic conditions). However, retirement is potentially endogenous with respect to cognition, since cognition may impact on decisions relating to retirement. If this is the case, the OLS estimates will be biased. In order to test for this possibility, instrumental variable (IV) estimation is used. This method requires an “instrumental variable” that is highly correlated with retirement duration but not correlated with cognition. The instrument used in this paper is based on the so-called "Marriage Bar", which was the legal requirement that women leave paid employment upon getting married. This law took effect in Ireland in the 1930s and was abolished only in the 1970s. The IV regression estimates, along with formal statistical tests, provide no evidence in support of the view that cognition impacts on retirement decisions. The finding of a small negative effect of retirement duration on cognition is shown to be robust to alternative empirical specifications. These findings are discussed in the wider context of the effects of work-like and work-related activities on cognition.