This dissertation aims to provide new insight into the properties that underlie scalar implicature (SI) processing. Previous experiments have investigated the time course of SI processing and whether or not it is a costly, resource-demanding process, but not what the resources have been specifically used to do, and when. Additionally, word scales used in SI research have been treated as interchangeable variations on a type, without systematic evaluation of that assumption. This dissertation investigates these questions via a computational model and several experiments using highly sensitive methodologies, including eyetracking. It concludes that scalar implicature is a costly, effortful process, though that effort is easily affected by task or other experimental factors. Specifically, processing difficulty appears to arise locally (as the scalar term is encountered) and the effort put towards a scalar inference, where the higher term(s) on the scale are negated. This negation appears to lead to reduced accessibility of the term.