diff --git a/doc/api/next_api_changes/behavior/hist2d_unit_conversion.rst b/doc/api/next_api_changes/behavior/hist2d_unit_conversion.rst
new file mode 100644
index 000000000000..56db226ceb47
--- /dev/null
+++ b/doc/api/next_api_changes/behavior/hist2d_unit_conversion.rst
@@ -0,0 +1,11 @@
+``Axes.hist2d`` now converts *x*/*y*/*range* through the axis unit converters
+------------------------------------------------------------------------------
+
+`~matplotlib.axes.Axes.hist2d` previously passed *x* and *y* directly to
+`numpy.histogram2d` without applying the axes' unit converters, unlike
+`~.Axes.plot`, `~.Axes.scatter`, and `~.Axes.hist`. This meant that, e.g.,
+plotting ``datetime64`` data with `~.Axes.hist2d` produced bin edges in raw
+(often nanosecond-scale) units instead of Matplotlib's internal date
+representation, so the histogram did not line up with other artists plotted
+on the same Axes. ``x``, ``y``, and ``range`` (if passed) are now converted
+consistently with the other plotting methods.
diff --git a/lib/matplotlib/axes/_axes.py b/lib/matplotlib/axes/_axes.py
index c71c260b9ed1..27be4babf62c 100644
--- a/lib/matplotlib/axes/_axes.py
+++ b/lib/matplotlib/axes/_axes.py
@@ -7992,8 +7992,20 @@ def hist2d(self, x, y, bins=10, range=None, density=False, weights=None,
            Previously, `~.Axes.hist2d` would force the axes limits to match the
            extents of the histogram; now, autoscaling also takes other plot
            elements into account.
+
+        .. versionchanged:: 3.12
+           *x* and *y* (and *range*, if given) are now passed through the axes'
+           unit converters before binning, so e.g. datetime inputs are handled
+           the same way as in `~.Axes.plot` and `~.Axes.scatter`. Previously
+           they were passed unconverted to `numpy.histogram2d`.
         """
 
+        x, y = self._process_unit_info([("x", x), ("y", y)], kwargs)
+        if range is not None:
+            (xmin, xmax), (ymin, ymax) = range
+            range = (self.convert_xunits((xmin, xmax)),
+                     self.convert_yunits((ymin, ymax)))
+
         h, xedges, yedges = np.histogram2d(x, y, bins=bins, range=range,
                                            density=density, weights=weights)
 
diff --git a/lib/matplotlib/tests/test_axes.py b/lib/matplotlib/tests/test_axes.py
index 968c41d61c02..abfa5aec86c8 100644
--- a/lib/matplotlib/tests/test_axes.py
+++ b/lib/matplotlib/tests/test_axes.py
@@ -2919,6 +2919,46 @@ def test_hist2d_autolimits():
     assert ax.get_autoscale_on()  # Autolimits have not been disabled.
 
 
+def test_hist2d_datetime():
+    # Regression test for gh-17319: hist2d's returned bin edges for
+    # datetime input should be on the same numeric scale as
+    # matplotlib.dates.date2num (i.e. equivalent to converting the dates
+    # yourself and calling np.histogram2d directly), not raw
+    # (nanosecond-scale) datetime64 integers.
+    x = np.arange(np.datetime64('2020-01-01'), np.datetime64('2020-01-11'))
+    y = np.arange(10.)
+
+    ax = plt.figure().add_subplot()
+    h, xedges, yedges, pc = ax.hist2d(x, y, bins=5)
+
+    expected_h, expected_xedges, expected_yedges = np.histogram2d(
+        mdates.date2num(x), y, bins=5)
+    np.testing.assert_array_equal(xedges, expected_xedges)
+    np.testing.assert_array_equal(yedges, expected_yedges)
+    np.testing.assert_array_equal(h, expected_h)
+
+
+def test_hist2d_datetime_range():
+    # The `range` parameter should be converted through the unit converters
+    # just like x/y, so passing datetime bounds is equivalent to manually
+    # converting them (e.g. via `date2num`) and passing the result.
+    x = np.arange(np.datetime64('2020-01-01'), np.datetime64('2020-01-11'))
+    y = np.arange(10.)
+    xlim_datetime = [np.datetime64('2020-01-01'), np.datetime64('2020-01-11')]
+    xlim_converted = mdates.date2num(xlim_datetime)
+
+    h_datetime, xedges_datetime, yedges_datetime, _ = (
+        plt.figure().add_subplot().hist2d(
+            x, y, bins=5, range=[xlim_datetime, [0, 10]]))
+    h_converted, xedges_converted, yedges_converted, _ = (
+        plt.figure().add_subplot().hist2d(
+            mdates.date2num(x), y, bins=5, range=[xlim_converted, [0, 10]]))
+
+    np.testing.assert_array_equal(xedges_datetime, xedges_converted)
+    np.testing.assert_array_equal(yedges_datetime, yedges_converted)
+    np.testing.assert_array_equal(h_datetime, h_converted)
+
+
 class TestScatter:
     @image_comparison(['scatter'], style='mpl20', remove_text=True)
     def test_scatter_plot(self):