diff --git a/reciprocalspaceship/dataset.py b/reciprocalspaceship/dataset.py
index 04c56f19..aa187a49 100644
--- a/reciprocalspaceship/dataset.py
+++ b/reciprocalspaceship/dataset.py
@@ -1177,7 +1177,7 @@ def unstack_anomalous(self, columns=None, suffixes=("(+)", "(-)")):
 
         return result
 
-    def is_isomorphous(self, other, cell_threshold=0.05):
+    def is_isomorphous(self, other, cell_threshold=0.5):
         """
         Determine whether DataSet is isomorphous to another DataSet. This
         method confirms isomorphism by ensuring the spacegroups are equivalent,
@@ -1214,7 +1214,8 @@ def is_isomorphous(self, other, cell_threshold=0.05):
         for param in params:
             param1 = self.cell.__getattribute__(param)
             param2 = other.cell.__getattribute__(param)
-            if (np.abs((param1 - param2)) / 100.0) > cell_threshold:
+            diff = 200.0 * np.abs(param1 - param2) / (param1 + param2)
+            if diff > cell_threshold:
                 return False
 
         return True
diff --git a/tests/test_dataset.py b/tests/test_dataset.py
index ec931d86..e0b4bf61 100644
--- a/tests/test_dataset.py
+++ b/tests/test_dataset.py
@@ -603,6 +603,48 @@ def test_is_isomorphous(data_unmerged, data_fmodel, sg1, sg2, cell1, cell2):
             assert not result
 
 
+@pytest.mark.parametrize("threshold", [5.0, 1.0, 0.5, 0.1])
+def test_is_isomorphous_threshold(threshold):
+    """
+    Test that DataSet.is_isorphous(self, other, cell_threshold) method's
+    cell_threshold operates on percent difference.
+    """
+    epsilon = 1e-12
+    cell = np.array([34.0, 45.0, 98.0, 90.0, 90.0, 90.0])
+    spacegroup = 19
+
+    ds = rs.DataSet(cell=cell, spacegroup=spacegroup)
+    cell_resize_factor = (200.0 + threshold) / (200.0 - threshold)
+
+    # Make a cell that should be exactly threshold percent bigger
+    other_cell = cell_resize_factor * cell
+    too_big_cell = other_cell + epsilon
+    big_cell = other_cell - epsilon
+
+    # Make a cell that should be exactly threshold percent smaller
+    other_cell = cell / cell_resize_factor
+    too_small_cell = other_cell - epsilon
+    small_cell = other_cell + epsilon
+
+    # Construct data sets
+    too_big = rs.DataSet(cell=too_big_cell, spacegroup=spacegroup)
+    big = rs.DataSet(cell=big_cell, spacegroup=spacegroup)
+    too_small = rs.DataSet(cell=too_small_cell, spacegroup=spacegroup)
+    small = rs.DataSet(cell=small_cell, spacegroup=spacegroup)
+
+    # Cell is barely too big to be isomorphous
+    assert not ds.is_isomorphous(too_big, threshold)
+
+    # Cell is barely too small to be isomorphous
+    assert not ds.is_isomorphous(too_small, threshold)
+
+    # Cell is almost too big to be isomorphous
+    assert ds.is_isomorphous(big, threshold)
+
+    # Cell is almost too small to be isomorphous
+    assert ds.is_isomorphous(small, threshold)
+
+
 def test_to_gemmi_withNans(data_merged):
     """
     GH144: Test whether DataSet.to_gemmi() works with NaN-containing data.