diff --git a/src/OpenColorIO/ops/lut1d/Lut1DOp.cpp b/src/OpenColorIO/ops/lut1d/Lut1DOp.cpp
index 412462414b..06fd2c19b4 100644
--- a/src/OpenColorIO/ops/lut1d/Lut1DOp.cpp
+++ b/src/OpenColorIO/ops/lut1d/Lut1DOp.cpp
@@ -16,6 +16,7 @@
 #include "ops/lut1d/Lut1DOpCPU.h"
 #include "ops/lut1d/Lut1DOpGPU.h"
 #include "ops/matrix/MatrixOp.h"
+#include "ops/range/RangeOp.h"
 #include "ops/OpTools.h"
 #include "SSE.h"
 #include "transforms/Lut1DTransform.h"
@@ -119,12 +120,35 @@ void Lut1DOp::combineWith(OpRcPtrVec & ops, ConstOpRcPtr & secondOp) const
     ConstLut1DOpRcPtr typedRcPtr = DynamicPtrCast<const Lut1DOp>(secondOp);
     auto secondLut = typedRcPtr->lut1DData();
 
-    // We want compose to upsample the LUTs to minimize precision loss.
-    const auto compFlag = Lut1DOpData::COMPOSE_RESAMPLE_BIG;
-    auto thisLut = lut1DData();
-    Lut1DOpDataRcPtr result =  Lut1DOpData::Compose(thisLut, secondLut, compFlag);
-    auto composedOp = std::make_shared<Lut1DOp>(result);
-    ops.push_back(composedOp);
+    // We are trying to combine a pair of forward/inverse LUTs.
+    // Use similair logic than OpOptimizers::RemoveInverseOps to replace the
+    // pair by a clamp where appropriate.
+    if (isInverse(secondOp))
+    {
+        auto opData = lut1DData()->getPairIdentityReplacement(secondLut);
+
+        if (opData->getType() == OpData::MatrixType)
+        {
+            // No-op that will be optimized.
+            auto mat = OCIO_DYNAMIC_POINTER_CAST<MatrixOpData>(opData);
+            CreateMatrixOp(ops, mat, TRANSFORM_DIR_FORWARD);
+        }
+        else if (opData->getType() == OpData::RangeType)
+        {
+            // Clamping op.
+            auto range = OCIO_DYNAMIC_POINTER_CAST<RangeOpData>(opData);
+            CreateRangeOp(ops, range, TRANSFORM_DIR_FORWARD);
+        }
+    }
+    else
+    {
+        // We want compose to upsample the LUTs to minimize precision loss.
+        const auto compFlag = Lut1DOpData::COMPOSE_RESAMPLE_BIG;
+        auto thisLut = lut1DData();
+        Lut1DOpDataRcPtr result =  Lut1DOpData::Compose(thisLut, secondLut, compFlag);
+        auto composedOp = std::make_shared<Lut1DOp>(result);
+        ops.push_back(composedOp);
+    }
 }
 
 bool Lut1DOp::hasChannelCrosstalk() const
diff --git a/tests/cpu/OpOptimizers_tests.cpp b/tests/cpu/OpOptimizers_tests.cpp
index 6f606fd6e8..077bbf623b 100644
--- a/tests/cpu/OpOptimizers_tests.cpp
+++ b/tests/cpu/OpOptimizers_tests.cpp
@@ -368,6 +368,32 @@ OCIO_ADD_TEST(OpOptimizers, combine_ops)
         OCIO::CombineOps(ops, OCIO::OPTIMIZATION_ALL);
         OCIO_CHECK_EQUAL(ops.size(), 0);
     }
+
+    {
+        // When a pair of forward / inverse LUTs with non 0 to 1 domain are used
+        // as process space for a Look (eg. CDL), the Optimizer tries to combine
+        // them when the Look results in a no-op. Here we make sure this result
+        // in an appropriate clamp instead of a new half-domain LUT resulting
+        // from the naive composition of the two LUTs.
+
+        OCIO::OpRcPtrVec ops;
+        const std::string fileName("lut1d_4.spi1d");
+        OCIO::ContextRcPtr context = OCIO::Context::Create();
+        OCIO_CHECK_NO_THROW(OCIO::BuildOpsTest(ops, fileName, context,
+                                            OCIO::TRANSFORM_DIR_INVERSE));
+        const double exp_null[4] = {1.0, 1.0, 1.0, 1.0};
+        OCIO::CreateExponentOp(ops, exp_null, OCIO::TRANSFORM_DIR_FORWARD);
+        OCIO_CHECK_NO_THROW(OCIO::BuildOpsTest(ops, fileName, context,
+                                            OCIO::TRANSFORM_DIR_FORWARD));
+
+        OCIO_CHECK_NO_THROW(ops.finalize());
+        OCIO_CHECK_NO_THROW(ops.optimize(OCIO::OPTIMIZATION_ALL));
+        OCIO_CHECK_EQUAL(ops.size(), 1);
+        OCIO::ConstOpRcPtr op = ops[0];
+        OCIO_REQUIRE_ASSERT(op);
+        auto range = OCIO_DYNAMIC_POINTER_CAST<const OCIO::RangeOpData>(op->data());
+        OCIO_REQUIRE_ASSERT(range);
+    }
 }
 
 OCIO_ADD_TEST(OpOptimizers, non_optimizable)