diff --git a/src/ArborX_LinearBVH.hpp b/src/ArborX_LinearBVH.hpp
index 9279f333b..a7b915b63 100644
--- a/src/ArborX_LinearBVH.hpp
+++ b/src/ArborX_LinearBVH.hpp
@@ -36,6 +36,12 @@
 namespace ArborX
 {
 
+namespace Experimental
+{
+struct PerThread
+{};
+} // namespace Experimental
+
 namespace Details
 {
 struct HappyTreeFriends;
@@ -101,6 +107,18 @@ class BasicBoundingVolumeHierarchy
         std::forward<View>(view), std::forward<Args>(args)...);
   }
 
+  template <typename Predicate, typename Callback>
+  KOKKOS_FUNCTION void query(Experimental::PerThread,
+                             Predicate const &predicate,
+                             Callback const &callback) const
+  {
+    ArborX::Details::TreeTraversal<BasicBoundingVolumeHierarchy,
+                                   /* Predicates Dummy */ std::true_type,
+                                   Callback, typename Predicate::Tag>
+        tree_traversal(*this, callback);
+    tree_traversal(predicate);
+  }
+
 private:
   friend struct Details::HappyTreeFriends;
 
@@ -161,6 +179,14 @@ class BoundingVolumeHierarchy
                      std::forward<CallbackOrView>(callback_or_view),
                      std::forward<View>(view), std::forward<Args>(args)...);
   }
+
+  template <typename Predicate, typename Callback>
+  KOKKOS_FUNCTION void query(Experimental::PerThread tag,
+                             Predicate const &predicate,
+                             Callback const &callback) const
+  {
+    base_type::query(tag, predicate, callback);
+  }
 };
 
 template <typename MemorySpace>
diff --git a/src/details/ArborX_DetailsTreeTraversal.hpp b/src/details/ArborX_DetailsTreeTraversal.hpp
index c94c2c5cb..ce758c6aa 100644
--- a/src/details/ArborX_DetailsTreeTraversal.hpp
+++ b/src/details/ArborX_DetailsTreeTraversal.hpp
@@ -63,15 +63,23 @@ struct TreeTraversal<BVH, Predicates, Callback, SpatialPredicateTag>
     else
     {
       Kokkos::parallel_for("ArborX::TreeTraversal::spatial",
-                           Kokkos::RangePolicy<ExecutionSpace>(
+                           Kokkos::RangePolicy<ExecutionSpace, FullTree>(
                                space, 0, Access::size(predicates)),
                            *this);
     }
   }
 
+  KOKKOS_FUNCTION TreeTraversal(BVH const &bvh, Callback const &callback)
+      : _bvh{bvh}
+      , _callback{callback}
+  {}
+
   struct OneLeafTree
   {};
 
+  struct FullTree
+  {};
+
   KOKKOS_FUNCTION void operator()(OneLeafTree, int queryIndex) const
   {
     auto const &predicate = Access::get(_predicates, queryIndex);
@@ -84,10 +92,15 @@ struct TreeTraversal<BVH, Predicates, Callback, SpatialPredicateTag>
     }
   }
 
-  KOKKOS_FUNCTION void operator()(int queryIndex) const
+  KOKKOS_FUNCTION void operator()(FullTree, int queryIndex) const
   {
     auto const &predicate = Access::get(_predicates, queryIndex);
+    operator()(predicate);
+  }
 
+  template <typename Predicate>
+  KOKKOS_FUNCTION void operator()(Predicate const &predicate) const
+  {
     int node = HappyTreeFriends::getRoot(_bvh); // start with root
     do
     {
@@ -415,15 +428,23 @@ struct TreeTraversal<BVH, Predicates, Callback,
     {
       Kokkos::parallel_for(
           "ArborX::Experimental::TreeTraversal::OrderedSpatialPredicate",
-          Kokkos::RangePolicy<ExecutionSpace>(space, 0,
-                                              Access::size(predicates)),
+          Kokkos::RangePolicy<ExecutionSpace, FullTree>(
+              space, 0, Access::size(predicates)),
           *this);
     }
   }
 
+  KOKKOS_FUNCTION TreeTraversal(BVH const &bvh, Callback const &callback)
+      : _bvh{bvh}
+      , _callback{callback}
+  {}
+
   struct OneLeafTree
   {};
 
+  struct FullTree
+  {};
+
   KOKKOS_FUNCTION void operator()(OneLeafTree, int queryIndex) const
   {
     auto const &predicate = Access::get(_predicates, queryIndex);
@@ -440,9 +461,15 @@ struct TreeTraversal<BVH, Predicates, Callback,
     }
   }
 
-  KOKKOS_FUNCTION void operator()(int queryIndex) const
+  KOKKOS_FUNCTION void operator()(FullTree, int queryIndex) const
   {
     auto const &predicate = Access::get(_predicates, queryIndex);
+    operator()(predicate);
+  }
+
+  template <typename Predicate>
+  KOKKOS_FUNCTION void operator()(Predicate const &predicate) const
+  {
     using ArborX::Details::HappyTreeFriends;
 
     using distance_type = decltype(predicate.distance(
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 85dbc5dfa..7afac2d56 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -124,6 +124,7 @@ foreach(_test Callbacks Degenerate ManufacturedSolution ComparisonWithBoost)
   endforeach()
 endforeach()
 list(APPEND ARBORX_TEST_QUERY_TREE_SOURCES
+  tstQueryTreeCallbackQueryPerThread.cpp
   tstQueryTreeRay.cpp
   tstQueryTreeTraversalPolicy.cpp
   tstQueryTreeIntersectsKDOP.cpp
diff --git a/test/tstQueryTreeCallbackQueryPerThread.cpp b/test/tstQueryTreeCallbackQueryPerThread.cpp
new file mode 100644
index 000000000..2734c4bf8
--- /dev/null
+++ b/test/tstQueryTreeCallbackQueryPerThread.cpp
@@ -0,0 +1,113 @@
+/****************************************************************************
+ * Copyright (c) 2023 by the ArborX authors                                 *
+ * All rights reserved.                                                     *
+ *                                                                          *
+ * This file is part of the ArborX library. ArborX is                       *
+ * distributed under a BSD 3-clause license. For the licensing terms see    *
+ * the LICENSE file in the top-level directory.                             *
+ *                                                                          *
+ * SPDX-License-Identifier: BSD-3-Clause                                    *
+ ****************************************************************************/
+
+#include "ArborX_EnableDeviceTypes.hpp" // ARBORX_DEVICE_TYPES
+#include <ArborX.hpp>
+
+#include "BoostTest_CUDA_clang_workarounds.hpp"
+#include <boost/test/unit_test.hpp>
+
+#include <numeric>
+#include <vector>
+
+BOOST_AUTO_TEST_SUITE(PerThread)
+
+struct IntersectionCallback
+{
+  int query_index;
+  bool &success;
+
+  template <typename Query, typename Value>
+  KOKKOS_FUNCTION void operator()(Query const &, Value const &value) const
+  {
+    success = (query_index == value.index);
+  }
+};
+
+BOOST_AUTO_TEST_CASE_TEMPLATE(callback_intersects, DeviceType,
+                              ARBORX_DEVICE_TYPES)
+{
+  using MemorySpace = typename DeviceType::memory_space;
+  using ExecutionSpace = typename DeviceType::execution_space;
+  using Tree = ArborX::BVH<MemorySpace>;
+
+  int const n = 10;
+  Kokkos::View<ArborX::Point *, DeviceType> points(
+      Kokkos::view_alloc(Kokkos::WithoutInitializing, "points"), n);
+  Kokkos::parallel_for(
+      Kokkos::RangePolicy<ExecutionSpace>(0, n), KOKKOS_LAMBDA(int i) {
+        points(i) = {{(float)i, (float)i, (float)i}};
+      });
+
+  Tree const tree(ExecutionSpace{}, points);
+
+  bool success;
+  Kokkos::parallel_reduce(
+      Kokkos::RangePolicy<ExecutionSpace>(0, n),
+      KOKKOS_LAMBDA(int i, bool &update) {
+        float center = i;
+        ArborX::Box box{{center - .5f, center - .5f, center - .5f},
+                        {center + .5f, center + .5f, center + .5f}};
+        tree.query(ArborX::Experimental::PerThread{}, ArborX::intersects(box),
+                   IntersectionCallback{i, update});
+      },
+      Kokkos::LAnd<bool, Kokkos::HostSpace>(success));
+
+  BOOST_TEST(success);
+}
+
+struct OrderedIntersectionCallback
+{
+  int query_index;
+  bool &success;
+
+  template <typename Query, typename Value>
+  KOKKOS_FUNCTION auto operator()(Query const &, Value const &value) const
+  {
+    success = (query_index == value.index);
+    return ArborX::CallbackTreeTraversalControl::early_exit;
+  }
+};
+
+BOOST_AUTO_TEST_CASE_TEMPLATE(callback_ordered_intersects, DeviceType,
+                              ARBORX_DEVICE_TYPES)
+{
+  using MemorySpace = typename DeviceType::memory_space;
+  using ExecutionSpace = typename DeviceType::execution_space;
+  using Tree = ArborX::BVH<MemorySpace>;
+
+  int const n = 10;
+  Kokkos::View<ArborX::Point *, DeviceType> points(
+      Kokkos::view_alloc(Kokkos::WithoutInitializing, "points"), n);
+  Kokkos::parallel_for(
+      Kokkos::RangePolicy<ExecutionSpace>(0, n), KOKKOS_LAMBDA(int i) {
+        points(i) = {{(float)i, (float)i, (float)i}};
+      });
+
+  Tree const tree(ExecutionSpace{}, points);
+
+  bool success;
+  Kokkos::parallel_reduce(
+      Kokkos::RangePolicy<ExecutionSpace>(0, n),
+      KOKKOS_LAMBDA(int i, bool &update) {
+        float center = i;
+        ArborX::Box box{{center - .5f, center - .5f, center - .5f},
+                        {center + .5f, center + .5f, center + .5f}};
+        tree.query(ArborX::Experimental::PerThread{},
+                   ArborX::Experimental::ordered_intersects(box),
+                   OrderedIntersectionCallback{i, update});
+      },
+      Kokkos::LAnd<bool, Kokkos::HostSpace>(success));
+
+  BOOST_TEST(success);
+}
+
+BOOST_AUTO_TEST_SUITE_END()