diff --git a/monai/apps/detection/transforms/dictionary.py b/monai/apps/detection/transforms/dictionary.py
index fc65954358..2c29937aa4 100644
--- a/monai/apps/detection/transforms/dictionary.py
+++ b/monai/apps/detection/transforms/dictionary.py
@@ -43,7 +43,7 @@
 from monai.transforms.transform import MapTransform, Randomizable, RandomizableTransform
 from monai.transforms.utils import generate_pos_neg_label_crop_centers, map_binary_to_indices
 from monai.utils import ImageMetaKey as Key
-from monai.utils import InterpolateMode, NumpyPadMode, ensure_tuple, ensure_tuple_rep
+from monai.utils import InterpolateMode, NumpyPadMode, ensure_tuple, ensure_tuple_rep, fall_back_tuple
 from monai.utils.enums import PostFix, TraceKeys
 from monai.utils.type_conversion import convert_data_type
 
@@ -1174,9 +1174,8 @@ def randomize(  # type: ignore
 
     def __call__(self, data: Mapping[Hashable, torch.Tensor]) -> List[Dict[Hashable, torch.Tensor]]:
         d = dict(data)
-        spatial_dims = len(d[self.image_keys[0]].shape) - 1
         image_size = d[self.image_keys[0]].shape[1:]
-        self.spatial_size = ensure_tuple_rep(self.spatial_size_, spatial_dims)
+        self.spatial_size = fall_back_tuple(self.spatial_size_, image_size)
 
         # randomly sample crop centers
         boxes = d[self.box_keys]
diff --git a/tests/test_box_transform.py b/tests/test_box_transform.py
index 6b0a4a2b19..ad82b080f1 100644
--- a/tests/test_box_transform.py
+++ b/tests/test_box_transform.py
@@ -16,7 +16,12 @@
 from parameterized import parameterized
 
 from monai.apps.detection.transforms.box_ops import convert_mask_to_box
-from monai.apps.detection.transforms.dictionary import BoxToMaskd, ConvertBoxModed, MaskToBoxd
+from monai.apps.detection.transforms.dictionary import (
+    BoxToMaskd,
+    ConvertBoxModed,
+    MaskToBoxd,
+    RandCropBoxByPosNegLabeld,
+)
 from monai.transforms import CastToTyped
 from tests.utils import TEST_NDARRAYS, assert_allclose
 
@@ -287,6 +292,24 @@ def test_value_3d(
             # assert_allclose(data_back["boxes"], data["boxes"], type_test=False, device_test=False, atol=1e-3)
             # assert_allclose(data_back["image"], data["image"], type_test=False, device_test=False, atol=1e-3)
 
+    def test_crop_shape(self):
+        tt = RandCropBoxByPosNegLabeld(
+            image_keys=["image"],
+            box_keys="box",
+            label_keys="label",
+            spatial_size=[10, 7, -1],
+            whole_box=True,
+            num_samples=1,
+            pos=1,
+            neg=0,
+        )
+        iii = {
+            "image": torch.rand(1, 10, 8, 7),
+            "box": torch.tensor(((1.0, 2.0, 3.0, 4.0, 5.0, 6.0),)),
+            "label": torch.tensor((1,)).long(),
+        }
+        self.assertEqual(tt(iii)[0]["image"].shape, (1, 10, 7, 7))
+
 
 if __name__ == "__main__":
     unittest.main()