diff --git a/openaddr/conform.py b/openaddr/conform.py
index 111486be..80f73e7a 100644
--- a/openaddr/conform.py
+++ b/openaddr/conform.py
@@ -716,18 +716,26 @@ def csv_source_to_csv(source_config, source_path, dest_path):
 def geojson_source_to_csv(source_config, source_path, dest_path):
     '''
     '''
+    # Not every feature shares the same set of properties, so make a first
+    # pass to collect the union of every feature's property keys (in
+    # first-seen order) before opening the CSV writer.
+    out_fieldnames = []
+    seen_fieldnames = set()
+    with open(source_path) as file:
+        for feature in stream_geojson(file):
+            for key in feature['properties'].keys():
+                if key not in seen_fieldnames:
+                    seen_fieldnames.add(key)
+                    out_fieldnames.append(key)
+    out_fieldnames.append(GEOM_FIELDNAME)
+
     # For every row in the source GeoJSON
     with open(source_path) as file:
         # Write the extracted CSV file
         with open(dest_path, 'w', encoding='utf-8') as dest_fp:
-            writer = None
+            writer = csv.DictWriter(dest_fp, out_fieldnames)
+            writer.writeheader()
             for (row_number, feature) in enumerate(stream_geojson(file)):
-                if writer is None:
-                    out_fieldnames = list(feature['properties'].keys())
-                    out_fieldnames.append(GEOM_FIELDNAME)
-                    writer = csv.DictWriter(dest_fp, out_fieldnames)
-                    writer.writeheader()
-
                 try:
                     row = feature['properties']
                     if feature['geometry'] is None:
diff --git a/openaddr/tests/conform.py b/openaddr/tests/conform.py
index f35ef831..bec50520 100644
--- a/openaddr/tests/conform.py
+++ b/openaddr/tests/conform.py
@@ -2212,6 +2212,47 @@ def test_geojson_source_to_csv(self):
             self.assertEqual(row[GEOM_FIELDNAME], 'POINT (-74.9833483425103 40.05498715)')
             self.assertEqual(row['PARCEL_NUM'], '02-022-003')
 
+    def test_geojson_source_to_csv_non_uniform_properties(self):
+        ''' Features with different property keys should not crash the writer.
+        '''
+        c = SourceConfig(dict({
+            "schema": 2,
+            "layers": {
+                "addresses": [{
+                    "name": "default",
+                    "conform": { }
+                }]
+            }
+        }), "addresses", "default")
+
+        geojson_path = os.path.join(self.testdir, 'non-uniform.geojson')
+        with open(geojson_path, 'w', encoding='utf8') as file:
+            json.dump({
+                "type": "FeatureCollection",
+                "features": [
+                    {
+                        "type": "Feature",
+                        "properties": {"pid": "1"},
+                        "geometry": {"type": "Point", "coordinates": [-121.2, 39.3]}
+                    },
+                    {
+                        "type": "Feature",
+                        "properties": {"pid": "2", "address": "123 Maple St"},
+                        "geometry": {"type": "Point", "coordinates": [-121.3, 39.4]}
+                    }
+                ]
+            }, file)
+
+        csv_path = os.path.join(self.testdir, 'non-uniform-conformed.csv')
+        geojson_source_to_csv(c, geojson_path, csv_path)
+
+        with open(csv_path, encoding='utf8') as file:
+            rows = list(csv.DictReader(file))
+            self.assertEqual(rows[0]['pid'], '1')
+            self.assertEqual(rows[0]['address'], '')
+            self.assertEqual(rows[1]['pid'], '2')
+            self.assertEqual(rows[1]['address'], '123 Maple St')
+
 class TestConformCsv(unittest.TestCase):
     "Fixture to create real files to test csv_source_to_csv()"