This article describes the methodology we have followed in order to improve the efficiency of a parsing algorithm for wide coverage unification-based grammars. The technique used is the restriction technique (Shieber 85), which has been recognized as an important operation to obtain efficient parsers for unification-based grammars. The main objective of the research is how to choose appropriate restrictors for using the restriction technique. We have developed a statistical model for selecting restrictors. Several experiments have been done in order to characterise those restrictors.