Mental health problems and severe cognitive decline can affect individual behaviours and physical health, cause adverse outcomes and generate significant economic costs for the society. In this paper we contribute to the existing literature by investigating long-term retirement effects on individual health outcomes for a developing country – China. Specifically, we examine the cumulative effect of years in retirement on cognitive functioning, measured by scores in cognitive tests such as word recall test and numeracy test, separately for men and women in China. Identifying such effects can be challenging due to endogeneity issues. To overcome this problem, we use different mandatory retirement ages as instruments for blue-collar and white-collar workers using three waves (2011-2015) of data from the national China Health and Retirement Longitudinal Study. Bivariate random-effect Tobit models are estimated, which also allows us to account for the left censored nature of retirement duration. We find that retirement duration is endogenous to men’s scores in memory tests. After accounting for the observed and unobserved confounding factors, we find that one additional year in retirement reduces the number of words men recalled immediately and 10 minutes later by 0.9% and 1.7%, respectively. The effects for women are much smaller - only 25-35% of those for men. One more year in retirement reduces men’ scores in numeracy test by 0.5%, and scores in mental intactness test by 0.3%, but the effect on women’s scores are statistically insignificant. We also explore the underlying mechanisms for these effects by examining participation in various physical and social activities post-retirement.