This paper investigates the bi-directional causal relationship between retirement and stress. We use PSID data for the period 2007-2015. Using a simultaneous equations approach, we find that a rise in stress increases the probability to retire by roughly 15.4 percentage points, while retirement decreases stress by 34.5 percentage points. We find the same results when we disaggregate by individuals' characteristics, but the former effect is stronger for males, for people working in typical blue-collar jobs, and for people whose wealth is below the mean; while the latter is stronger for males, for white-collar workers, for people whose wealth is above the mean, and for white individuals. We show that official retirement ages are a strong instrument for actual retirement age, and that lagged physical activity levels are a non-linear instrument for perceived stress. We also confirm that objective measures of mental health are a strong instrument for perceived stress.