In this study we use five biennial panels of males in the Retirement History survey to estimate a model of self assessed health status and retirement status, both of which are categorical dependent variables whose residuals may be correlated. The health variable can take one of four states ranging from better health than others the same age to dead. The retirement variable has two states: working or not working full time. Our maximum likelihood estimator requires ten-way integration of the multivariate normal density function. We outline the Gaussian quadrature formulae used in calculating these multivariate distribution functions. We find many significant variables. Perhaps of most interest is the trade off in retirement between Social Security benefits and wage rates and the sharp decline in retirement before age 62, the earliest date nondisabled men are eligible for benefits. We also find that health is worse for those whose longest occupation is unskilled labor.