We discuss a bottom-up parsing technique based on delayed reductions, and investigate its capabilities and limitations. Some non-LR(k) grammars, for any k, are handled deterministically by this method. Surprisingly, and counter-intuitively from the viewpoint of LR(k), increase of delay may lead to decrease of determinism. We also present a variant that uses both delay and lookahead. Key words: Formal languages, grammars, parsing, LR(k)