In this paper we present a notion of expansion of a term in the lambda-calculus which transforms terms into linear terms. This transformation replaces each occurrence of a variable in the original term by a fresh variable taking into account non-trivial implications in the structure of the term caused by these simple replacements. We prove that the class of terms which can be expanded is the same of terms typable in an Intersection Type System, i.e. the strongly normalizable terms. We then show that expansion is preserved by weak-head reduction, the reduction considered by functional programming languages.