diff --git a/CHANGES.rst b/CHANGES.rst
index 251cb20f..281070b8 100644
--- a/CHANGES.rst
+++ b/CHANGES.rst
@@ -37,6 +37,9 @@ Bug Fixes
 - Fix ``rebuild_fits_rec_dtype`` handling of unsigned integer columns
   with shapes [#213]
 
+- Fix unit roundtripping when writing to a datamodel with a table
+  to a FITS file [#242]
+
 Changes to API
 --------------
 
diff --git a/src/stdatamodels/jwst/datamodels/tests/test_fits.py b/src/stdatamodels/jwst/datamodels/tests/test_fits.py
index 4ba9bb96..5f54b041 100644
--- a/src/stdatamodels/jwst/datamodels/tests/test_fits.py
+++ b/src/stdatamodels/jwst/datamodels/tests/test_fits.py
@@ -5,7 +5,7 @@
 import pytest
 
 from stdatamodels.jwst import datamodels
-from stdatamodels.jwst.datamodels import ImageModel, JwstDataModel, RampModel
+from stdatamodels.jwst.datamodels import ImageModel, JwstDataModel, RampModel, SpecModel
 
 
 @pytest.fixture
@@ -91,3 +91,18 @@ def test_resave_duplication_bug(tmp_path):
 
     with fits.open(fn1) as ff1, fits.open(fn2) as ff2:
         assert ff1['ASDF'].size == ff2['ASDF'].size
+
+
+def test_units_roundtrip(tmp_path):
+    m = SpecModel()
+    # this next line is required for stdatamodels to cast
+    # spec_table to a FITS_rec (similar to having data assigned
+    # to the attribute)
+    m.spec_table = m.spec_table
+    m.spec_table.columns['WAVELENGTH'].unit = 'nm'
+
+    fn = tmp_path / "test1.fits"
+    m.save(fn)
+
+    m = datamodels.open(fn)
+    assert m.spec_table.columns['WAVELENGTH'].unit == 'nm'
diff --git a/src/stdatamodels/properties.py b/src/stdatamodels/properties.py
index 113c8e00..caf44f42 100644
--- a/src/stdatamodels/properties.py
+++ b/src/stdatamodels/properties.py
@@ -83,10 +83,20 @@ def _cast(val, schema):
                     t['shape'] = shape
 
         dtype = ndarray.asdf_datatype_to_numpy_dtype(schema['datatype'])
+
+        # save columns in case this is cast back to a fitsrec
+        if hasattr(val, 'columns'):
+            cols = val.columns
+        else:
+            cols = None
         val = util.gentle_asarray(val, dtype, allow_extra_columns=allow_extra_columns)
 
         if dtype.fields is not None:
             val = _as_fitsrec(val)
+            if cols is not None:
+                for col in cols:
+                    if col.name in val.names and col.unit is not None:
+                        val.columns[col.name].unit = col.unit
 
     if 'ndim' in schema and len(val.shape) != schema['ndim']:
         raise ValueError(