I present a dynamic structural discrete choice model of dairy cattle replacement to examine culling decisions and explain why there is a discrepancy between the optimal culling rates predicted in previous economic models and the replacement behaviors observed in practice. The empirical model estimates the parameters of the dairy manager’s payoff function using a dataset of cow-level milk test and replacement records on a New York dairy farm, and is one of the few dairy cattle replacement models to incorporate within-lactation variations. The results suggest that a rational dairy manager would take into account the opportunity cost of keeping milk cows longer, which reduces his valuation of the lactating cows, and speeds up the replacement to retain the economic value of milk cows at the upper level over time.