diff --git a/src/biocutils/normalize_subscript.py b/src/biocutils/normalize_subscript.py
index 3514149..fa699d7 100644
--- a/src/biocutils/normalize_subscript.py
+++ b/src/biocutils/normalize_subscript.py
@@ -13,7 +13,7 @@ def _raise_int(idx: int, length):
     pass
 
 
-def normalize_subscript(sub: Union[slice, range, Sequence, int, str, bool], length: int, names: Optional[Sequence[str]] = None) -> Tuple:
+def normalize_subscript(sub: Union[slice, range, Sequence, int, str, bool], length: int, names: Optional[Sequence[str]] = None, non_negative_only: bool = True) -> Tuple:
     """
     Normalize a subscript for ``__getitem__`` or friends into a sequence of
     integer indices, for consistent downstream use.
@@ -45,6 +45,10 @@ def normalize_subscript(sub: Union[slice, range, Sequence, int, str, bool], leng
             List of names for each entry in the object. If not None, this
             should have length equal to ``length``.
 
+        non_negative_only:
+            Whether negative indices must be converted into non-negative
+            equivalents. Setting this to `False` may improve efficiency.
+
     Returns:
         A tuple containing (i) a sequence of integer indices in ``[0, length)``
         specifying the subscript elements, and (ii) a boolean indicating whether
@@ -59,7 +63,7 @@ def normalize_subscript(sub: Union[slice, range, Sequence, int, str, bool], leng
     if isinstance(sub, int) or (has_numpy and isinstance(sub, numpy.generic)):
         if sub < -length or sub >= length:
             _raise_int(sub, length)
-        if sub < 0:
+        if sub < 0 and non_negative_only:
             sub += length
         return [int(sub)], True
 
@@ -85,26 +89,29 @@ def normalize_subscript(sub: Union[slice, range, Sequence, int, str, bool], leng
         if last < -length:
             _raise_int(last, length)
 
-        if sub.start < 0:
-            if sub.stop < 0:
-                return range(length + sub.start, length + sub.stop, sub.step), False
-            else:
-                return [ (x < 0) * length + x for x in sub], False
+        if not non_negative_only:
+            return sub, False
         else:
-            if sub.stop < 0:
-                return [ (x < 0) * length + x for x in sub], False
+            if sub.start < 0:
+                if sub.stop < 0:
+                    return range(length + sub.start, length + sub.stop, sub.step), False
+                else:
+                    return [ (x < 0) * length + x for x in sub], False
             else:
-                return sub, False
+                if sub.stop < 0:
+                    return [ (x < 0) * length + x for x in sub], False
+                else:
+                    return sub, False
 
-    can_return_early = False
+    can_return_early = True
     for x in sub:
-        if isinstance(x, str) or isinstance(x, bool) or (has_numpy and isinstance(x, numpy.bool_)) or x < 0:
+        if isinstance(x, str) or isinstance(x, bool) or (has_numpy and isinstance(x, numpy.bool_)) or (x < 0 and non_negative_only):
             can_return_early = False;
             break
 
     if can_return_early:
         for x in sub:
-            if x >= length:
+            if x >= length or x < -length:
                 _raise_int(x, length)
         return sub, False
 
@@ -122,11 +129,11 @@ def normalize_subscript(sub: Union[slice, range, Sequence, int, str, bool], leng
         elif x < 0:
             if x < -length:
                 _raise_int(x, length)
-            output.append(x + length)
+            output.append(int(x) + length)
         else:
             if x >= length:
                 _raise_int(x, length)
-            output.append(x)
+            output.append(int(x))
 
     if len(has_strings):
         if names is None:
diff --git a/tests/test_normalize_subscript.py b/tests/test_normalize_subscript.py
index a6fc02d..3516ae7 100644
--- a/tests/test_normalize_subscript.py
+++ b/tests/test_normalize_subscript.py
@@ -99,3 +99,14 @@ def test_normalize_subscript_numpy():
 
     # Now the trickiest part - are booleans converted correctly?
     assert normalize_subscript(numpy.array([True, False, True, False, True]), 5) == ([0, 2, 4], False)
+
+
+def test_normalize_subscript_allow_negative():
+    assert normalize_subscript(-50, 100, non_negative_only=False) == ([-50], True)
+    assert normalize_subscript(range(50, -10, -1), 100, non_negative_only=False) == (range(50, -10, -1), False)
+    assert normalize_subscript(range(-10, -50, -1), 100, non_negative_only=False) == (range(-10, -50, -1), False)
+    assert normalize_subscript([0,-1,2,-3,4,-5,6,-7,8], 50, non_negative_only=False) == ([0,-1,2,-3,4,-5,6,-7,8], False)
+
+    with pytest.raises(IndexError) as ex:
+        normalize_subscript([0,-1,2,-3,4,-51,6,-7,8], 50, non_negative_only=False)
+    assert str(ex.value).find("subscript (-51) out of range") >= 0