It is a common and acceptable practice to present a preliminary version of a paper at a conference and then to submit the full paper to a refereed journal. The Foreword of a recent STOC proceedings states, for example,
The submissions were not refereed, and many of these papers represent reports of continuing research. It is expected that most of them will appear in a more polished and complete form in scientific journals.That is the ideal. In practice, many of us have been in the habit of putting complete versions of accepted papers in conference proceedings. This is not a wise strategy. Assuming that the publisher of the conference proceedings holds copyright of the papers that appears in the proceedings, the paper that one submits to a refereed journal must be substantially different from the one that appears in a conference proceeding in order to avoid copyright infringement. I asked my contact at Springer, the publisher of the journal Theory of Computing Systems, how different the journal version of a paper should be from the conference version, and was told to use 25% as an acceptable guideline. This is a reasonable response that should not too difficult to manage for theorists, and can be taken up by proofs and explanations that were not included in the conference version. I should add however, that Lance Fortnow asked the same question of IEEE, the copyright holder for papers that appear in the Conference on Computational Complexity, and the response was far more onerous. Lance's contact at IEEE called for 30% to 70% new material, and material from the original paper should be there only to support new ideas. I would rather abide by Springer's recommendation than IEEE's. The essential point is that the journal version must be essentially different from the conference version.