Trust dilution and trust fusion are two operators that are used to calculate transitive trust in a trust network. Various implementations of these operators already exist but are not fully motivated. In this paper we define the basic properties of these two operators by developing a trust algebra. We evaluate several new and existing models against the axioms of this algebra, amongst which a number of variations of the Subjective Logic. The algebra enables the comparison of models and gives more insight in the available recommendation models and their properties.