This paper argues for a distinction between word formation fromroots and word formation from existing words. Focusing on Hebrew, it is shown that roots – and only roots – may be assigned multiple interpretations in different environments. When the basisfor the derivation is a word, this word forces its semantic and phonological properties on any element derived from it. To account for this difference, a locality constraint on the interpretation of roots is postulated: the first nominal or verbal head that merges with the root serves as the immediate environment for determining its interpretation. This head forms a closed domain:any further derivation takes as its input not the root itself, but an element whose semantic and phonological properties have been cashed out. Word-derived words thus have access only to the words they are derived from, not to the root. While the ability of Hebrew roots to acquiremultiple interpretations is language specific, the distinction between word formation from roots and word formation from words is shown to be universal. This is illustrated here with Englishzero-related pairs, which are shown to exhibit the same contrasts as Hebrew between word formation from roots and from words. Showing the effect of roots in word formation in both Hebrew and English further motivates the root hypothesis, namely, that in all languages the lexical kernel,or the root, is distinct from `words' – complex entities – even if this distinction is not always morphologically manifested.