diff --git a/changes/8958.general.rst b/changes/8958.general.rst
new file mode 100644
index 0000000000..a84befe1ff
--- /dev/null
+++ b/changes/8958.general.rst
@@ -0,0 +1 @@
+When blending metadata don't store columns containing all missing value (nans).
diff --git a/jwst/model_blender/_tablebuilder.py b/jwst/model_blender/_tablebuilder.py
index e677d0699e..7bfd2b4efb 100644
--- a/jwst/model_blender/_tablebuilder.py
+++ b/jwst/model_blender/_tablebuilder.py
@@ -1,6 +1,12 @@
 import numpy as np
 
 
+class _MissingValueType:
+    pass
+
+_MISSING_VALUE = _MissingValueType()
+
+
 def _convert_dtype(value):
     """Convert numarray column dtype into YAML-compatible format description"""
     if 'U' in value:
@@ -99,7 +105,7 @@ def header_to_row(self, header):
 
     def _add_row(self, row):
         for attr, col in self.attr_to_column.items():
-            self.columns[col].append(row[attr] if attr in row else np.nan)
+            self.columns[col].append(row[attr] if attr in row else _MISSING_VALUE)
 
     def build_table(self):
         """
@@ -113,6 +119,8 @@ def build_table(self):
         arrays = []
         table_dtype = []
         for col, items in self.columns.items():
-            arrays.append(np.array(items))
+            if all((i is _MISSING_VALUE for i in items)):
+                continue
+            arrays.append(np.array([np.nan if i is _MISSING_VALUE else i for i in items]))
             table_dtype.append((col, arrays[-1].dtype))
         return np.rec.fromarrays(arrays, dtype=table_dtype)
diff --git a/jwst/model_blender/tests/test_blend.py b/jwst/model_blender/tests/test_blend.py
index 37c9d705dc..48f0fbd891 100644
--- a/jwst/model_blender/tests/test_blend.py
+++ b/jwst/model_blender/tests/test_blend.py
@@ -37,6 +37,11 @@
 # even if the metadata is defined in the schema
 KNOWN_MISSING_FIRSTS = [None, '1', '2']
 
+# None of the below test data defines this attribute.
+# It is expected to be missing from the resulting
+# table (as is checked below).
+MISSING_COLUMN = "TIME-OBS"
+
 
 def _make_data():
     """Create a set of input models to blendmeta
@@ -195,6 +200,7 @@ def test_blendtab(blend):
     # Ensure all the expected FITS keywords are in the table.
     colnames = set(newtab.dtype.fields)
     assert not fits_expected.difference(colnames)
+    assert MISSING_COLUMN not in colnames
     for col in colnames:
         if col in input_values:
             assert newtab[col] == input_values[col]