diff --git a/lib/matplotlib/legend.py b/lib/matplotlib/legend.py
index 4f18d0526e6..3af482cf603 100644
--- a/lib/matplotlib/legend.py
+++ b/lib/matplotlib/legend.py
@@ -737,6 +737,7 @@ def _auto_legend_data(self):
         ax = self.parent
         bboxes = []
         lines = []
+        offsets = []
 
         for handle in ax.lines:
             assert isinstance(handle, Line2D)
@@ -755,12 +756,19 @@ def _auto_legend_data(self):
                 transform = handle.get_transform()
                 bboxes.append(handle.get_path().get_extents(transform))
 
+        for handle in ax.collections:
+            transform, transOffset, hoffsets, paths = handle._prepare_points()
+
+            if len(hoffsets):
+                for offset in transOffset.transform(hoffsets):
+                    offsets.append(offset)
+
         try:
             vertices = np.concatenate([l.vertices for l in lines])
         except ValueError:
             vertices = np.array([])
 
-        return [vertices, bboxes, lines]
+        return [vertices, bboxes, lines, offsets]
 
     def draw_frame(self, b):
         'b is a boolean.  Set draw frame to b'
@@ -920,7 +928,7 @@ def _find_best_position(self, width, height, renderer, consider=None):
         # should always hold because function is only called internally
         assert self.isaxes
 
-        verts, bboxes, lines = self._auto_legend_data()
+        verts, bboxes, lines, offsets = self._auto_legend_data()
 
         bbox = Bbox.from_bounds(0, 0, width, height)
         if consider is None:
@@ -939,6 +947,7 @@ def _find_best_position(self, width, height, renderer, consider=None):
             # take their into account when checking vertex overlaps in
             # the next line.
             badness = legendBox.count_contains(verts)
+            badness += legendBox.count_contains(offsets)
             badness += legendBox.count_overlaps(bboxes)
             for line in lines:
                 # FIXME: the following line is ill-suited for lines
diff --git a/lib/matplotlib/tests/baseline_images/test_legend/not_covering_scatter.png b/lib/matplotlib/tests/baseline_images/test_legend/not_covering_scatter.png
new file mode 100644
index 00000000000..4e9b72e640c
Binary files /dev/null and b/lib/matplotlib/tests/baseline_images/test_legend/not_covering_scatter.png differ
diff --git a/lib/matplotlib/tests/baseline_images/test_legend/not_covering_scatter_transform.png b/lib/matplotlib/tests/baseline_images/test_legend/not_covering_scatter_transform.png
new file mode 100644
index 00000000000..e2a90732fc1
Binary files /dev/null and b/lib/matplotlib/tests/baseline_images/test_legend/not_covering_scatter_transform.png differ
diff --git a/lib/matplotlib/tests/test_legend.py b/lib/matplotlib/tests/test_legend.py
index a43231b964e..293c64a3dbf 100644
--- a/lib/matplotlib/tests/test_legend.py
+++ b/lib/matplotlib/tests/test_legend.py
@@ -16,6 +16,7 @@
 import matplotlib.pyplot as plt
 import matplotlib as mpl
 import matplotlib.patches as mpatches
+import matplotlib.transforms as mtrans
 
 
 @image_comparison(baseline_images=['legend_auto1'], remove_text=True)
@@ -261,6 +262,32 @@ def test_nanscatter():
     ax.grid(True)
 
 
+@image_comparison(baseline_images=['not_covering_scatter'], extensions=['png'])
+def test_not_covering_scatter():
+    colors = ['b','g','r']
+
+    for n in range(3):
+        plt.scatter([n,], [n,], color=colors[n])
+
+    plt.legend(['foo', 'foo', 'foo'], loc='best')
+    plt.gca().set_xlim(-0.5, 2.2)
+    plt.gca().set_ylim(-0.5, 2.2)
+
+
+@image_comparison(baseline_images=['not_covering_scatter_transform'],
+                  extensions=['png'])
+def test_not_covering_scatter_transform():
+    # Offsets point to top left, the default auto position
+    offset = mtrans.Affine2D().translate(-20, 20)
+    x = np.linspace(0, 30, 1000)
+    plt.plot(x, x)
+
+    plt.scatter([20], [10], transform=offset + plt.gca().transData)
+
+    plt.legend(['foo', 'bar'], loc='best')
+
+
 if __name__ == '__main__':
     import nose
     nose.runmodule(argv=['-s', '--with-doctest'], exit=False)
+