diff --git a/opm/simulators/flow/FlowGenericProblem_impl.hpp b/opm/simulators/flow/FlowGenericProblem_impl.hpp
index 9be32dacf5..7beb6d616f 100644
--- a/opm/simulators/flow/FlowGenericProblem_impl.hpp
+++ b/opm/simulators/flow/FlowGenericProblem_impl.hpp
@@ -339,12 +339,17 @@ typename FlowGenericProblem<GridView,FluidSystem>::Scalar
 FlowGenericProblem<GridView,FluidSystem>::
 rockFraction(unsigned elementIdx, unsigned timeIdx) const
 {
-    // the reference porosity is defined as the accumulated pore volume divided by the
-    // geometric volume of the element. Note that it can
-    // be larger than 1.0 if porevolume multipliers are used
-    // to for instance implement larger boundary cells
-    auto porosity = this->lookUpData_.fieldPropDouble(eclState_.fieldProps(), "PORO", elementIdx);
-    return referencePorosity(elementIdx, timeIdx) / porosity * (1 - porosity);
+    // For the energy equation, we need the volume of the rock.
+    // The volume of the rock is computed by rockFraction * geometric volume of the element.
+    // The reference porosity is defined as porosity * ntg * pore-volume-multiplier.
+    // A common practice in reservoir simulation is to use large pore-volume-multipliers in boundary cells
+    // to model boundary conditions other than no-flow. This may result in reference porosities that are larger than 1.
+    // A simple (1-reference porosity) * geometric volume of the element may give unphysical results.
+    // We therefore instead consider the pore-volume-multiplier as a volume multiplier. The rock fraction is thus given by
+    // (1 - porosity * ntg) * pore-volume-multiplier = (1 - porosity * ntg) * reference porosity / (porosity * ntg)
+    const auto ntg = this->lookUpData_.fieldPropDouble(eclState_.fieldProps(), "NTG", elementIdx);
+    const auto poro_eff = ntg * this->lookUpData_.fieldPropDouble(eclState_.fieldProps(), "PORO", elementIdx);
+    return (1 - poro_eff) * referencePorosity(elementIdx, timeIdx) / poro_eff;
 }
 
 template<class GridView, class FluidSystem>