diff --git a/holoviews/plotting/plot.py b/holoviews/plotting/plot.py
index e57e9a7637..ab4add2572 100644
--- a/holoviews/plotting/plot.py
+++ b/holoviews/plotting/plot.py
@@ -705,11 +705,11 @@ def _compute_group_range(cls, group, elements, ranges, framewise,
                     data_range = el.range(el_dim, dimension_range=False)
 
                 data_ranges[(el, el_dim)] = data_range
-                if dtype is not None and dtype.kind == 'uif' and robust:
+                if dtype is not None and dtype.kind in 'uif' and robust:
                     percentile = 2 if isinstance(robust, bool) else robust
                     robust_ranges[(el, el_dim)] = (
                         dim(el_dim, np.nanpercentile, percentile).apply(el),
-                        dim(el_dim, np.nanpercentile, percentile).apply(el)
+                        dim(el_dim, np.nanpercentile, 100 - percentile).apply(el)
                     )
 
                 if (any(isinstance(r, str) for r in data_range) or
diff --git a/holoviews/tests/plotting/bokeh/test_elementplot.py b/holoviews/tests/plotting/bokeh/test_elementplot.py
index 88a30d30b2..19dbfaf215 100644
--- a/holoviews/tests/plotting/bokeh/test_elementplot.py
+++ b/holoviews/tests/plotting/bokeh/test_elementplot.py
@@ -971,3 +971,14 @@ def test_categorical_overlay_dimension_values_skip_factor(self):
         plot = bokeh_renderer.get_plot((curve*scatter).redim.values(x=['A', 'C']))
         x_range = plot.handles['x_range']
         self.assertEqual(x_range.factors, ['A', 'C'])
+
+    def test_clim_percentile(self):
+        arr = np.random.rand(10,10)
+        arr[0, 0] = -100
+        arr[-1, -1] = 100
+        im = Image(arr).opts(clim_percentile=True)
+
+        plot = bokeh_renderer.get_plot(im)
+        low, high = plot.ranges[('Image',)]['z']['robust']
+        assert low > 0
+        assert high < 1