This paper evaluates and compares the performance of two approaches for locating an agent in a mobile agent environment. The first approach dynamically creates a chain of forwarders to locate a moving agent whereas the second one relies on a centralized server to perform this task. Based on a Markov chain analysis, we compute the performance of each scheme (time to reach an agent, number of forwarders) and compare them first with simulations and second with experimental results obtained by using ProActive, a Java library. Depending on the system parameters we identify the best scheme and observe that over a LAN the server yields the best performance whereas the forwarders yield the best performance over a MAN. Key words: Mobile code, migration, centralized server, forwarders, Markov chain.