diff --git a/test/test-pad.c b/test/test-pad.c
index 5bce7e06..5e4a66f5 100644
--- a/test/test-pad.c
+++ b/test/test-pad.c
@@ -270,6 +270,9 @@ START_TEST(pad_ring)
 	struct libinput_event_tablet_pad *pev;
 	int val;
 	double degrees, expected;
+	int min, max;
+	int step_size;
+	int nevents = 0;
 
 	litest_pad_ring_start(dev, 10);
 
@@ -278,11 +281,28 @@ START_TEST(pad_ring)
 	/* Wacom's 0 value is at 275 degrees */
 	expected = 270;
 
-	for (val = 0; val < 100; val += 10) {
+	min = libevdev_get_abs_minimum(dev->evdev, ABS_WHEEL);
+	max = libevdev_get_abs_maximum(dev->evdev, ABS_WHEEL);
+	step_size = 360/(max - min + 1);
+
+	/* This is a bit strange because we rely on kernel filtering here.
+	   The litest_*() functions take a percentage, but mapping this to
+	   the pads 72 or 36 range pad ranges is lossy and a bit
+	   unpredictable. So instead we increase by a small percentage,
+	   expecting *most* events to be filtered by the kernel because they
+	   resolve to the same integer value as the previous event. Whenever
+	   an event gets through, we expect that to be the next integer
+	   value in the range and thus the next step on the circle.
+	 */
+	for (val = 0; val < 100.0; val += 1) {
 		litest_pad_ring_change(dev, val);
 		libinput_dispatch(li);
 
 		ev = libinput_get_event(li);
+		if (!ev)
+			continue;
+
+		nevents++;
 		pev = litest_is_pad_ring_event(ev,
 					       0,
 					       LIBINPUT_TABLET_PAD_RING_SOURCE_FINGER);
@@ -291,15 +311,14 @@ START_TEST(pad_ring)
 		ck_assert_double_ge(degrees, 0.0);
 		ck_assert_double_lt(degrees, 360.0);
 
-		/* rounding errors, mostly caused by small physical range */
-		ck_assert_double_ge(degrees, expected - 2);
-		ck_assert_double_le(degrees, expected + 2);
+		ck_assert_double_eq(degrees, expected);
 
 		libinput_event_destroy(ev);
-
-		expected = fmod(degrees + 36, 360);
+		expected = fmod(degrees + step_size, 360);
 	}
 
+	ck_assert_int_eq(nevents, 360/step_size - 1);
+
 	litest_pad_ring_end(dev);
 }
 END_TEST