diff --git a/django/db/models/base.py b/django/db/models/base.py
index a3478445f7..9bbac65385 100644
--- a/django/db/models/base.py
+++ b/django/db/models/base.py
@@ -334,8 +334,8 @@ class Model(object):
             qn(self._meta.db_table), qn(self._meta.pk.column), op)
         param = smart_str(getattr(self, field.attname))
         q = self.__class__._default_manager.filter(**kwargs).order_by((not is_next and '-' or '') + field.name, (not is_next and '-' or '') + self._meta.pk.name)
-        q._where.append(where)
-        q._params.extend([param, param, getattr(self, self._meta.pk.attname)])
+        q.extra(where=where, params=[param, param,
+            getattr(self, self._meta.pk.attname)])
         try:
             return q[0]
         except IndexError:
diff --git a/django/db/models/sql/query.py b/django/db/models/sql/query.py
index 6302416b41..eaa5f11d35 100644
--- a/django/db/models/sql/query.py
+++ b/django/db/models/sql/query.py
@@ -10,7 +10,7 @@ all about the internals of models in order to get the information it needs.
 import copy
 
 from django.utils import tree
-from django.db.models.sql.where import WhereNode, AND
+from django.db.models.sql.where import WhereNode, AND, OR
 from django.db.models.sql.datastructures import Count
 from django.db.models.fields import FieldDoesNotExist
 from django.contrib.contenttypes import generic
@@ -233,14 +233,27 @@ class Query(object):
         # Work out how to relabel the rhs aliases, if necessary.
         change_map = {}
         used = {}
+        first_new_join = True
         for alias in rhs.tables:
+            if not rhs.alias_map[alias][ALIAS_REFCOUNT]:
+                # An unused alias.
+                continue
             promote = (rhs.alias_map[alias][ALIAS_JOIN][JOIN_TYPE] ==
                     self.LOUTER)
             new_alias = self.join(rhs.rev_join_map[alias], exclusions=used,
-                    promote=promote)
+                    promote=promote, outer_if_first=True)
+            if self.alias_map[alias][ALIAS_REFCOUNT] == 1:
+                first_new_join = False
             used[new_alias] = None
             change_map[alias] = new_alias
 
+        # So that we don't exclude valid results, the first join that is
+        # exclusive to the lhs (self) must be converted to an outer join.
+        for alias in self.tables[1:]:
+            if self.alias_map[alias][ALIAS_REFCOUNT] == 1:
+                self.alias_map[alias][ALIAS_JOIN][JOIN_TYPE] = self.LOUTER
+                break
+
         # Now relabel a copy of the rhs where-clause and add it to the current
         # one.
         if rhs.where:
@@ -380,8 +393,12 @@ class Query(object):
         """ Decreases the reference count for this alias. """
         self.alias_map[alias][ALIAS_REFCOUNT] -= 1
 
+    def promote_alias(self, alias):
+        """ Promotes the join type of an alias to an outer join. """
+        self.alias_map[alias][ALIAS_JOIN][JOIN_TYPE] = self.LOUTER
+
     def join(self, (lhs, table, lhs_col, col), always_create=False,
-            exclusions=(), promote=False):
+            exclusions=(), promote=False, outer_if_first=False):
         """
         Returns an alias for a join between 'table' and 'lhs' on the given
         columns, either reusing an existing alias for that join or creating a
@@ -398,6 +415,10 @@ class Query(object):
         If 'promote' is True, the join type for the alias will be LOUTER (if
         the alias previously existed, the join type will be promoted from INNER
         to LOUTER, if necessary).
+
+        If 'outer_if_first' is True and a new join is created, it will have the
+        LOUTER join type. This is used when joining certain types of querysets
+        and Q-objects together.
         """
         if lhs not in self.alias_map:
             lhs_table = lhs
@@ -422,7 +443,7 @@ class Query(object):
         assert not is_table, \
                 "Must pass in lhs alias when creating a new join."
         alias, _ = self.table_alias(table, True)
-        join_type = promote and self.LOUTER or self.INNER
+        join_type = (promote or outer_if_first) and self.LOUTER or self.INNER
         join = [table, alias, join_type, lhs, lhs_col, col]
         if not lhs:
             # Not all tables need to be joined to anything. No join type
@@ -487,7 +508,8 @@ class Query(object):
         opts = self.model._meta
         alias = self.join((None, opts.db_table, None, None))
         dupe_multis = (connection == AND)
-        last = None
+        seen_aliases = []
+        done_split = not self.where
 
         # FIXME: Using enumerate() here is expensive. We only need 'i' to
         # check we aren't joining against a non-joinable field. Find a
@@ -498,23 +520,35 @@ class Query(object):
             if name == 'pk':
                 name = target_field.name
             if joins is not None:
+                seen_aliases.extend(joins)
                 last = joins
                 alias = joins[-1]
+                if connection == OR and not done_split:
+                    if self.alias_map[joins[0]][ALIAS_REFCOUNT] == 1:
+                        done_split = True
+                        self.promote_alias(joins[0])
+                        for t in self.tables[1:]:
+                            if t in seen_aliases:
+                                continue
+                            self.promote_alias(t)
+                            break
+                    else:
+                        seen_aliases.extend(joins)
             else:
                 # Normal field lookup must be the last field in the filter.
                 if i != len(parts) - 1:
-                    raise TypeError("Joins on field %r not permitted."
+                    raise TypeError("Join on field %r not permitted."
                             % name)
 
         col = target_col or target_field.column
 
-        if target_field is opts.pk and last:
+        if target_field is opts.pk and seen_aliases:
             # An optimization: if the final join is against a primary key,
             # we can go back one step in the join chain and compare against
             # the lhs of the join instead. The result (potentially) involves
             # one less table join.
             self.unref_alias(alias)
-            join = self.alias_map[last[-1]][ALIAS_JOIN]
+            join = self.alias_map[seen_aliases[-1]][ALIAS_JOIN]
             alias = join[LHS_ALIAS]
             col = join[LHS_JOIN_COL]
 
@@ -523,13 +557,13 @@ class Query(object):
             # join when connecting to the previous model. We make that
             # adjustment here. We don't do this unless needed because it's less
             # efficient at the database level.
-            self.alias_map[joins[0]][ALIAS_JOIN][JOIN_TYPE] = self.LOUTER
+            self.promote_alias(joins[0])
 
         self.where.add([alias, col, orig_field, lookup_type, value],
                 connection)
         if negate:
-            if last:
-                self.alias_map[last[0]][ALIAS_JOIN][JOIN_TYPE] = self.LOUTER
+            if seen_aliases:
+                self.promote_alias(last[0])
             self.where.negate()
 
     def add_q(self, q_object):
diff --git a/django/db/models/sql/where.py b/django/db/models/sql/where.py
index 5995f83486..527ce833fc 100644
--- a/django/db/models/sql/where.py
+++ b/django/db/models/sql/where.py
@@ -64,11 +64,19 @@ class WhereNode(tree.Node):
                 else:
                     format = '(%s)'
             else:
-                sql = self.make_atom(child)
-                params = child[2].get_db_prep_lookup(child[3], child[4])
-                format = '%s'
-            result.append(format % sql)
-            result_params.extend(params)
+                try:
+                    sql = self.make_atom(child)
+                    params = child[2].get_db_prep_lookup(child[3], child[4])
+                    format = '%s'
+                except EmptyResultSet:
+                    if node.negated:
+                        # If this is a "not" atom, being empty means it has no
+                        # effect on the result, so we can ignore it.
+                        continue
+                    raise
+            if sql:
+                result.append(format % sql)
+                result_params.extend(params)
         conn = ' %s ' % node.connection
         return conn.join(result), result_params
 
diff --git a/django/utils/tree.py b/django/utils/tree.py
index 0f83c38a73..0dc09e83a2 100644
--- a/django/utils/tree.py
+++ b/django/utils/tree.py
@@ -21,6 +21,10 @@ class Node(object):
         self.subtree_parents = []
         self.negated = False
 
+    def __str__(self):
+        return '(%s: %s)' % (self.connection, ', '.join([str(c) for c in
+            self.children]))
+
     def __deepcopy__(self, memodict):
         """
         Utility method used by copy.deepcopy().
@@ -59,7 +63,8 @@ class Node(object):
         if len(self.children) < 2:
             self.connection = conn_type
         if self.connection == conn_type:
-            if isinstance(node, Node) and node.connection == conn_type:
+            if isinstance(node, Node) and (node.connection == conn_type
+                    or len(node) == 1):
                 self.children.extend(node.children)
             else:
                 self.children.append(node)
diff --git a/tests/modeltests/lookup/models.py b/tests/modeltests/lookup/models.py
index aa90d1e5ec..06070b9a01 100644
--- a/tests/modeltests/lookup/models.py
+++ b/tests/modeltests/lookup/models.py
@@ -258,7 +258,7 @@ TypeError: Cannot resolve keyword 'pub_date_year' into field. Choices are: id, h
 >>> Article.objects.filter(headline__starts='Article')
 Traceback (most recent call last):
     ...
-TypeError: Cannot resolve keyword 'headline__starts' into field. Choices are: id, headline, pub_date
+TypeError: Join on field 'headline' not permitted.
 
 # Create some articles with a bit more interesting headlines for testing field lookups:
 >>> now = datetime.now()
diff --git a/tests/regressiontests/queries/models.py b/tests/regressiontests/queries/models.py
index 38e89dc515..8a24160419 100644
--- a/tests/regressiontests/queries/models.py
+++ b/tests/regressiontests/queries/models.py
@@ -111,11 +111,9 @@ Bug #4464
 >>> Item.objects.filter(tags__in=[t1, t2]).filter(tags=t3)
 [<Item: two>]
 
-Bug #2080
-# FIXME: Still problematic: the join needs to be "left outer" on the reverse
-# fk, but the individual joins only need to be inner.
-# >>> Author.objects.filter(Q(name='a3') | Q(item__name='one'))
-# [<Author: a3>]
+Bug #2080, #3592
+>>> Author.objects.filter(Q(name='a3') | Q(item__name='one'))
+[<Author: a1>, <Author: a3>]
 
 Bug #2939
 # FIXME: ValueQuerySets don't work yet.