Following the approach of Tversky's contrast model, similarity measures are introduced as order-preserving maps from specific ordered sets into the non-negative reals. The analysis of this class of functions gives rise to the definition of additive similarity measures that induce a matroid on their domain. An explicit representation of this matroid is given that can be used to directly determine the additive extension of a map from the basis of the matroid. In addition, a criteria is given to check, whether an additive map is a similarity measure or not.