Bounded Model Checking, although complete in theory, has been thus far limited in practice to falsification of properties that were not invariants. In this paper we propose a termination criterion for all of LTL, and we show its effectiveness through experiments. Our approach is based on converting the LTL formula to a B¨uchi automaton so as to reduce model checking to the verification of a fairness constraint. This reduction leads to one termination criterion that applies to all formulae. We also discuss cases for which a dedicated termination test improves bounded model checking efficiency.