A predictor of a context-free grammar G is a substring of a sentence in $L(G)$ which determines unambiguously the contents of the parse stack immediately before (in top-down parsing) or after (in bottom-up parsing) symbols of the predictor are processed. Two types of predictors are defined, one for bottom-up parsers and the other for top-down parsers. Algorithms for finding predictors are given and the possible applications of predictors are discussed.