diff --git a/bson/_cbsonmodule.c b/bson/_cbsonmodule.c
index 5c5d721ec..f00c6f3c5 100644
--- a/bson/_cbsonmodule.c
+++ b/bson/_cbsonmodule.c
@@ -1424,17 +1424,34 @@ int write_dict(PyObject* self, buffer_t buffer,
     }
 
     /* Write _id first if this is a top level doc. */
-    if (top_level && PyMapping_HasKeyString(dict, "_id")) {
-        PyObject* _id = PyMapping_GetItemString(dict, "_id");
-        if (!_id) {
-            return 0;
-        }
-        if (!write_pair(self, buffer, "_id", 3,
-                        _id, check_keys, options, 1)) {
+    if (top_level) {
+        /*
+         * If "dict" is a defaultdict we don't want to call
+         * PyMapping_GetItemString on it. That would **create**
+         * an _id where one didn't previously exist (PYTHON-871).
+         */
+        if (PyDict_Check(dict)) {
+            /* PyDict_GetItemString returns a borrowed reference. */
+            PyObject* _id = PyDict_GetItemString(dict, "_id");
+            if (_id) {
+                if (!write_pair(self, buffer, "_id", 3,
+                                _id, check_keys, options, 1)) {
+                    return 0;
+                }
+            }
+        } else if (PyMapping_HasKeyString(dict, "_id")) {
+            PyObject* _id = PyMapping_GetItemString(dict, "_id");
+            if (!_id) {
+                return 0;
+            }
+            if (!write_pair(self, buffer, "_id", 3,
+                            _id, check_keys, options, 1)) {
+                Py_DECREF(_id);
+                return 0;
+            }
+            /* PyMapping_GetItemString returns a new reference. */
             Py_DECREF(_id);
-            return 0;
         }
-        Py_DECREF(_id);
     }
 
     iter = PyObject_GetIter(dict);
diff --git a/test/test_bson.py b/test/test_bson.py
index 81ba22fce..b4b676c5c 100644
--- a/test/test_bson.py
+++ b/test/test_bson.py
@@ -143,6 +143,11 @@ class TestBSON(unittest.TestCase):
     def test_encode_then_decode_any_mapping(self):
         self.check_encode_then_decode(doc_class=NotADict)
 
+    def test_encoding_defaultdict(self):
+        dct = collections.defaultdict(dict, [('foo', 'bar')])
+        BSON.encode(dct)
+        self.assertEqual(dct, collections.defaultdict(dict, [('foo', 'bar')]))
+
     def test_basic_validation(self):
         self.assertRaises(TypeError, is_valid, 100)
         self.assertRaises(TypeError, is_valid, u("test"))
diff --git a/test/test_collection.py b/test/test_collection.py
index ebae73f54..7303e7b58 100644
--- a/test/test_collection.py
+++ b/test/test_collection.py
@@ -20,6 +20,8 @@ import re
 import sys
 import threading
 
+from collections import defaultdict
+
 sys.path[0:0] = [""]
 
 from bson.regex import Regex
@@ -657,6 +659,13 @@ class TestCollection(IntegrationTest):
         self.assertFalse(result.acknowledged)
         wait_until(lambda: 0 == db.test.count(), 'delete 2 documents')
 
+    def test_find_by_default_dct(self):
+        db = self.db
+        db.test.insert_one({'foo': 'bar'})
+        dct = defaultdict(dict, [('foo', 'bar')])
+        self.assertIsNotNone(db.test.find_one(dct))
+        self.assertEqual(dct, defaultdict(dict, [('foo', 'bar')]))
+
     def test_find_w_fields(self):
         db = self.db
         db.test.delete_many({})