A frequently invoked principle for theory appraisal in the sciences and humanities (including, especially, philosophy) are the so-called ‘virtues of a good theory’, such as simplicity, predictive power, and coherence with well-established theories. It is commonly held, though, that a clear understanding of the virtues is still lacking. An approach that has not yet been sufficiently appreciated as an answer to this problem has been pursued by Paul Thagard under the label ‘coherence-based inference’. As a specific implementation of this approach, Thagard’s ‘Theory of Explanatory Coherence’ (TEC) operationalizes the theoretical virtues and their interplay through formal ‘coherence principles’. However, TEC faces several challenges: (1) It is ambiguous with regard to the case of comparing more than two competing theories. (2) A preference for simpler (i.e., less ad hoc) explanations is not covered sufficiently in the coherence principles laid down within TEC. (3) A preference for more consilient (or ‘unified’) explanations is not sufficiently operationalized in the coherence principles. I propose solutions to challenges (1)-(3) and argue, on basis of a review of the theoretical virtues discussed in the literature, that the revised version of TEC can afford an illuminating model of theory appraisal, unifying the variety of virtues.