Recent research has addressed the problem of planning in non-deterministic domains. Classical planning has also been extended to the case of goals that can express temporal properties. However, the combination of these two aspects is not trivial. In non-deterministic domains, goals should take into account the fact that a plan may result in many possible different executions and that some requirements can be enforced on all the possible executions, while others may be enforced only on some executions. In this paper we address this problem. We define a planning algorithm that generates automatically plans for extended goals in nondeterministic domains. We also provide preliminary experimental results based on an implementation of the planning algorithm that uses symbolic model checking techniques.