diff --git a/pykernels/regular.py b/pykernels/regular.py
index c65f01e..ab48c93 100644
--- a/pykernels/regular.py
+++ b/pykernels/regular.py
@@ -465,16 +465,11 @@ def _compute(self, data_1, data_2):
         if np.any(data_1 < 0) or np.any(data_2 < 0):
             warnings.warn('MinMax kernel requires data to be strictly positive!')
 
-        minkernel = np.zeros((data_1.shape[0], data_2.shape[0]))
-        maxkernel = np.zeros((data_1.shape[0], data_2.shape[0]))
-
-        for d in range(data_1.shape[1]):
-            column_1 = data_1[:, d].reshape(-1, 1)
-            column_2 = data_2[:, d].reshape(-1, 1)
-            minkernel += np.minimum(column_1, column_2.T) 
-            maxkernel += np.maximum(column_1, column_2.T)
-
-        return minkernel/maxkernel
+        data_1 = data_1[:, None, :]
+        data_2 = data_2[None, :, :]
+        minkernel = np.minimum(data_1, data_2).sum(axis=2)  # Sum over features
+        maxkernel = np.maximum(data_1, data_2).sum(axis=2)
+        return minkernel / maxkernel
 
     def dim(self):
         return None