diff --git a/tests/test_error.py b/tests/test_error.py
index 5de66250..0de974be 100644
--- a/tests/test_error.py
+++ b/tests/test_error.py
@@ -5,6 +5,7 @@
 import pytest
 from ase.atoms import Atoms
 from ase.calculators.lj import LennardJones
+from ase.stress import voigt_6_to_full_3x3_stress
 from pytest import approx
 
 from pprint import pprint
@@ -103,6 +104,17 @@ def test_err_from_calc(ref_atoms):
     assert ref_err_dict['virial/atom/comp']['_ALL_']["count"] == 10 * 6
 
 
+def test_err_stress_shape(ref_atoms):
+    ref_atoms_calc = generic_calc(ref_atoms, OutputSpec(), LennardJones(sigma=0.75), output_prefix='calc_')
+    ref_err_dict, _, _ = ref_err_calc(ref_atoms_calc, ref_property_prefix='REF_', calc_property_prefix='calc_')
+
+    for at in ref_atoms_calc:
+        at.info["REF_stress"] = voigt_6_to_full_3x3_stress(at.info["REF_stress"])
+    ref_err_dict_shape, _, _ = ref_err_calc(ref_atoms_calc, ref_property_prefix='REF_', calc_property_prefix='calc_')
+
+    assert ref_err_dict == ref_err_dict_shape
+
+
 def test_error_properties(ref_atoms):
     ref_atoms_calc = generic_calc(ref_atoms, OutputSpec(), LennardJones(sigma=0.75), output_prefix='calc_')
     # both energy and per atom
diff --git a/wfl/fit/error.py b/wfl/fit/error.py
index e2986d64..14d97285 100755
--- a/wfl/fit/error.py
+++ b/wfl/fit/error.py
@@ -85,8 +85,6 @@ def _reshape_normalize(quant, prop, atoms, per_atom):
         quant: 2-d array containing reshaped quantity, with leading dimension 1 for per-config
             or len(atoms) for per-atom
         """
-        # convert scalars or lists into arrays
-        quant = np.asarray(quant)
 
         # fix shape of stress/virial
         if prop.startswith("stress") or prop.startswith("virial"):
@@ -156,9 +154,9 @@ def _reshape_normalize(quant, prop, atoms, per_atom):
                     raise ValueError("/atom only possible in config_properties")
                 data = at.arrays
 
-            # grab data
-            ref_quant = data.get(ref_property_prefix + prop_use)
-            calc_quant = data.get(calc_property_prefix + prop_use)
+            # grab data, make a copy so normalization doesn't affect original
+            ref_quant = np.asarray(data.get(ref_property_prefix + prop_use)).copy()
+            calc_quant = np.asarray(data.get(calc_property_prefix + prop_use)).copy()
             if ref_quant is None or calc_quant is None:
                 # warn if data is missing by reporting summary at the very end
                 if prop not in missed_prop_counter: