The preferred tool in theoretical dairy cow replacement studies has been stochastic dynamic programming. Under practical conditions, however, this method is less suitable due to computational complexity. In order to develop a ranking criterion for practical use, two criteria were tested by means of stochastic simulation based on random number generation. Under the first criterion the ranking was provided by a dynamic programming model, but instead of herd individual conditions a set of standard conditions was used. Under the second criterion the cows were ranked according to their expected net returns to housing, labour and management during the next 12 months. Both criteria were tested in the presence and absence of a milk quota. The simulation results showed that both criteria are suitable for practical implementation in both situations. Thus the final choice depends on other considerations including implementation and operation costs as well as comprehensibility. A third criterion called expected maximum net returns, which has been suggested in the literature, was considered, but, rejected from theoretical and empirical reasons.