Circumscription is a paradigm of non-monotonic logic meant to formalize the common-sense understanding that, among competing theories that represent phenomena equally well, the one with the fewest “abnormal” assumptions should be selected. Several papers have considered ways of adding circumscription to Description Logics. One of the proposals with good computational properties is Grounded Circumscription, introduced by Sengupta, Krishnadi and Hitzler in 2011. Our paper builds on their general idea, but identifies some problems with the original semantics definition, which gives rise to counter-intuitive consequences and renders the proposed tableau algorithm incorrect. We give an example that makes the problem explicit and propose a modification of the semantics that remedies this issue. On the algorithmic side, we show that a big part of the reasoning can actually be transferred to standard Description Logics, for which tools and results already exist.