diff --git a/lib/matplotlib/projections/polar.py b/lib/matplotlib/projections/polar.py
index 3fbb4d2b9cfd..3243c76d8661 100644
--- a/lib/matplotlib/projections/polar.py
+++ b/lib/matplotlib/projections/polar.py
@@ -16,13 +16,18 @@
 
 
 class PolarTransform(mtransforms.Transform):
-    """
+    r"""
     The base polar transform.
 
-    This transform maps polar coordinates ``(theta, r)`` into Cartesian
-    coordinates ``(x, y) = (r * cos(theta), r * sin(theta))`` (but does not
+    This transform maps polar coordinates :math:`\theta, r` into Cartesian
+    coordinates :math:`x, y = r \cos(\theta), r \sin(\theta)`
+    (but does not fully transform into Axes coordinates or
     handle positioning in screen space).
 
+    This transformation is designed to be applied to data after any scaling
+    along the radial axis (e.g. log-scaling) has been applied to the input
+    data.
+
     Path segments at a fixed radius are automatically transformed to circular
     arcs as long as ``path._interpolation_steps > 1``.
     """
@@ -30,7 +35,7 @@ class PolarTransform(mtransforms.Transform):
     input_dims = output_dims = 2
 
     def __init__(self, axis=None, use_rmin=True,
-                 _apply_theta_transforms=True):
+                 _apply_theta_transforms=True, *, scale_transform=None):
         """
         Parameters
         ----------
@@ -46,12 +51,18 @@ def __init__(self, axis=None, use_rmin=True,
         self._axis = axis
         self._use_rmin = use_rmin
         self._apply_theta_transforms = _apply_theta_transforms
+        self._scale_transform = scale_transform
 
     __str__ = mtransforms._make_str_method(
         "_axis",
         use_rmin="_use_rmin",
         _apply_theta_transforms="_apply_theta_transforms")
 
+    def _get_rorigin(self):
+        # Get lower r limit after being scaled by the radial scale transform
+        return self._scale_transform.transform(
+            (0, self._axis.get_rorigin()))[1]
+
     def transform_non_affine(self, tr):
         # docstring inherited
         theta, r = np.transpose(tr)
@@ -61,7 +72,7 @@ def transform_non_affine(self, tr):
             theta *= self._axis.get_theta_direction()
             theta += self._axis.get_theta_offset()
         if self._use_rmin and self._axis is not None:
-            r = (r - self._axis.get_rorigin()) * self._axis.get_rsign()
+            r = (r - self._get_rorigin()) * self._axis.get_rsign()
         r = np.where(r >= 0, r, np.nan)
         return np.column_stack([r * np.cos(theta), r * np.sin(theta)])
 
@@ -85,7 +96,7 @@ def transform_path_non_affine(self, path):
                     # that behavior here.
                     last_td, td = np.rad2deg([last_t, t])
                     if self._use_rmin and self._axis is not None:
-                        r = ((r - self._axis.get_rorigin())
+                        r = ((r - self._get_rorigin())
                              * self._axis.get_rsign())
                     if last_td <= td:
                         while td - last_td > 360:
@@ -877,7 +888,9 @@ def _set_lim_and_transforms(self):
         # data.  This one is aware of rmin
         self.transProjection = self.PolarTransform(
             self,
-            _apply_theta_transforms=False)
+            _apply_theta_transforms=False,
+            scale_transform=self.transScale
+        )
         # Add dependency on rorigin.
         self.transProjection.set_children(self._originViewLim)
 
@@ -888,9 +901,25 @@ def _set_lim_and_transforms(self):
 
         # The complete data transformation stack -- from data all the
         # way to display coordinates
+        #
+        # 1. Remove any radial axis scaling (e.g. log scaling)
+        # 2. Shift data in the theta direction
+        # 3. Project the data from polar to cartesian values
+        #    (with the origin in the same place)
+        # 4. Scale and translate the cartesian values to Axes coordinates
+        #    (here the origin is moved to the lower left of the Axes)
+        # 5. Move and scale to fill the Axes
+        # 6. Convert from Axes coordinates to Figure coordinates
         self.transData = (
-            self.transScale + self.transShift + self.transProjection +
-            (self.transProjectionAffine + self.transWedge + self.transAxes))
+            self.transScale +
+            self.transShift +
+            self.transProjection +
+            (
+                self.transProjectionAffine +
+                self.transWedge +
+                self.transAxes
+            )
+        )
 
         # This is the transform for theta-axis ticks.  It is
         # equivalent to transData, except it always puts r == 0.0 and r == 1.0
diff --git a/lib/matplotlib/tests/baseline_images/test_polar/polar_log.png b/lib/matplotlib/tests/baseline_images/test_polar/polar_log.png
new file mode 100644
index 000000000000..ab8e20b482a5
Binary files /dev/null and b/lib/matplotlib/tests/baseline_images/test_polar/polar_log.png differ
diff --git a/lib/matplotlib/tests/test_polar.py b/lib/matplotlib/tests/test_polar.py
index 1bde5bcc1142..1f8e6a75baca 100644
--- a/lib/matplotlib/tests/test_polar.py
+++ b/lib/matplotlib/tests/test_polar.py
@@ -434,3 +434,15 @@ def test_cursor_precision():
     assert ax.format_coord(2, 0) == "θ=0.6π (115°), r=0.000"
     assert ax.format_coord(2, .1) == "θ=0.64π (115°), r=0.100"
     assert ax.format_coord(2, 1) == "θ=0.637π (114.6°), r=1.000"
+
+
+@image_comparison(['polar_log.png'], style='default')
+def test_polar_log():
+    fig = plt.figure()
+    ax = fig.add_subplot(polar=True)
+
+    ax.set_rscale('log')
+    ax.set_rlim(1, 1000)
+
+    n = 100
+    ax.plot(np.linspace(0, 2 * np.pi, n), np.logspace(0, 2, n))