diff --git a/django/db/backends/oracle/query.py b/django/db/backends/oracle/query.py
index 8563420d7f..22ef1359a3 100644
--- a/django/db/backends/oracle/query.py
+++ b/django/db/backends/oracle/query.py
@@ -6,6 +6,7 @@ Derives from: django.db.models.sql.query.Query
 import datetime
 
 from django.db.backends import util
+from django.utils.encoding import force_unicode
 
 # Cache. Maps default query class to new Oracle query class.
 _classes = {}
@@ -55,6 +56,9 @@ def query_class(QueryClass, Database):
         def convert_values(self, value, field):
             if isinstance(value, Database.LOB):
                 value = value.read()
+                if field and field.get_internal_type() == 'TextField':
+                    value = force_unicode(value)
+                    
             # Oracle stores empty strings as null. We need to undo this in
             # order to adhere to the Django convention of using the empty
             # string instead of null, but only if the field accepts the
diff --git a/tests/regressiontests/datatypes/models.py b/tests/regressiontests/datatypes/models.py
index 7e76d6a669..0ad809a45d 100644
--- a/tests/regressiontests/datatypes/models.py
+++ b/tests/regressiontests/datatypes/models.py
@@ -13,6 +13,7 @@ class Donut(models.Model):
     baked_date = models.DateField(null=True)
     baked_time = models.TimeField(null=True)
     consumed_at = models.DateTimeField(null=True)
+    review = models.TextField()
 
     class Meta:
         ordering = ('consumed_at',)
@@ -82,6 +83,12 @@ datetime.datetime(2007, 4, 20, 16, 19, 59)
 >>> Donut.objects.filter(consumed_at__year=2008)
 []
 
+# Regression test for #10238: TextField values returned from the database
+# should be unicode.
+>>> d2 = Donut.objects.create(name=u'Jelly Donut', review=u'Outstanding')
+>>> Donut.objects.get(id=d2.id).review
+u'Outstanding'
+
 """}
 
 # Regression test for #8354: the MySQL backend should raise an error if given