We consider a Markov chain on a locally compact separable metric space $X$ and with a unique invariant probability. We show that such a chain can be classified into two categories according to the type of convergence of the expected occupation measures. Several properties in each category are investigated.