diff --git a/lib/matplotlib/transforms.py b/lib/matplotlib/transforms.py
index 2ac69726b005..cbeb4b037481 100644
--- a/lib/matplotlib/transforms.py
+++ b/lib/matplotlib/transforms.py
@@ -682,6 +682,82 @@ def intersection(bbox1, bbox2):
 class Bbox(BboxBase):
     """
     A mutable bounding box.
+
+    Examples
+    --------
+    **Create from known bounds**
+
+    The default constructor takes the boundary "points" ``[[xmin, ymin],
+    [xmax, ymax]]``.
+
+        >>> Bbox([[1, 1], [3, 7]])
+        Bbox([[1.0, 1.0], [3.0, 7.0]])
+
+    Alternatively, a Bbox can be created from the flattened points array, the
+    so-called "extents" ``(xmin, ymin, xmax, ymax)``
+
+        >>> Bbox.from_extents(1, 1, 3, 7)
+        Bbox([[1.0, 1.0], [3.0, 7.0]])
+
+    or from the "bounds" ``(xmin, ymin, width, height)``.
+
+        >>> Bbox.from_bounds(1, 1, 2, 6)
+        Bbox([[1.0, 1.0], [3.0, 7.0]])
+
+    **Create from collections of points**
+
+    The "empty" object for accumulating Bboxs is the null bbox, which is a
+    stand-in for the empty set.
+
+        >>> Bbox.null()
+        Bbox([[inf, inf], [-inf, -inf]])
+
+    Adding points to the null bbox will give you the bbox of those points.
+
+        >>> box = Bbox.null()
+        >>> box.update_from_data_xy([[1, 1]])
+        >>> box
+        Bbox([[1.0, 1.0], [1.0, 1.0]])
+        >>> box.update_from_data_xy([[2, 3], [3, 2]], ignore=False)
+        >>> box
+        Bbox([[1.0, 1.0], [3.0, 3.0]])
+
+    Setting ``ignore=True`` is equivalent to starting over from a null bbox.
+
+        >>> box.update_from_data_xy([[1, 1]], ignore=True)
+        >>> box
+        Bbox([[1.0, 1.0], [1.0, 1.0]])
+
+    .. warning::
+
+        It is recommended to always specify ``ignore`` explicitly.  If not, the
+        default value of ``ignore`` can be changed at any time by code with
+        access to your Bbox, for example using the method `~.Bbox.ignore`.
+
+    **Properties of the ``null`` bbox**
+
+    .. note::
+
+        The current behavior of `Bbox.null()` may be surprising as it does
+        not have all of the properties of the "empty set", and as such does
+        not behave like a "zero" object in the mathematical sense. We may
+        change that in the future (with a deprecation period).
+
+    The null bbox is the identity for intersections
+
+        >>> Bbox.intersection(Bbox([[1, 1], [3, 7]]), Bbox.null())
+        Bbox([[1.0, 1.0], [3.0, 7.0]])
+
+    except with itself, where it returns the full space.
+
+        >>> Bbox.intersection(Bbox.null(), Bbox.null())
+        Bbox([[-inf, -inf], [inf, inf]])
+
+    A union containing null will always return the full space (not the other
+    set!)
+
+        >>> Bbox.union([Bbox([[0, 0], [0, 0]]), Bbox.null()])
+        Bbox([[-inf, -inf], [inf, inf]])
     """
 
     def __init__(self, points, **kwargs):
@@ -690,12 +766,6 @@ def __init__(self, points, **kwargs):
         ----------
         points : ndarray
             A 2x2 numpy array of the form ``[[x0, y0], [x1, y1]]``.
-
-        Notes
-        -----
-        If you need to create a `Bbox` object from another form
-        of data, consider the static methods :meth:`unit`,
-        :meth:`from_bounds` and :meth:`from_extents`.
         """
         BboxBase.__init__(self, **kwargs)
         points = np.asarray(points, float)