diff --git a/toolkit.py b/toolkit.py
index 9a0a445826615792eddf0000c3231ac5b1cc0468..8d368613e7c8c34b38772b856e0a634e4533185c 100755
--- a/toolkit.py
+++ b/toolkit.py
@@ -828,13 +828,19 @@ class ClassificationProject(object):
         hist, bins = np.histogram(x, **np_kwargs)
         centers = (bins[:-1] + bins[1:]) / 2
         if "weights" in np_kwargs:
-            errors = []
-            for left, right in zip(bins, bins[1:]):
-                indices = np.where((x >= left) & (x < right))[0]
-                sumw2 = np.sum(np_kwargs["weights"][indices]**2)
-                content = np.sum(np_kwargs["weights"][indices])
-                errors.append(math.sqrt(sumw2)/content)
-            errors = np.array(errors)
+            bin_indices = np.digitize(x, bins)
+            sumw2 = np.array([np.sum(np_kwargs["weights"][bin_indices==i]**2)
+                              for i in range(1, len(bins)+1)])
+            sumw = np.array([np.sum(np_kwargs["weights"][bin_indices==i])
+                             for i in range(1, len(bins)+1)])
+            # move overflow to last bin
+            # (since thats what np.histogram gives us)
+            sumw2[-2] += sumw2[-1]
+            sumw2 = sumw2[:-1]
+            sumw[-2] += sumw[-1]
+            sumw = sumw[:-1]
+            # calculate relative error
+            errors = np.sqrt(sumw2)/sumw
         else:
             errors = np.sqrt(hist)/hist
         if scale_factor is not None: