diff --git a/rl_coach/architectures/tensorflow_components/heads/q_head.py b/rl_coach/architectures/tensorflow_components/heads/q_head.py
index 32c6946..eedec5b 100644
--- a/rl_coach/architectures/tensorflow_components/heads/q_head.py
+++ b/rl_coach/architectures/tensorflow_components/heads/q_head.py
@@ -34,6 +34,12 @@ class QHead(Head):
             self.num_actions = 1
         elif isinstance(self.spaces.action, DiscreteActionSpace):
             self.num_actions = len(self.spaces.action.actions)
+        else:
+            raise ValueError(
+                'QHead does not support action spaces of type: {class_name}'.format(
+                    class_name=self.spaces.action.__class__.__name__,
+                )
+            )
         self.return_type = QActionStateValue
         if agent_parameters.network_wrappers[self.network_name].replace_mse_with_huber_loss:
             self.loss_type = tf.losses.huber_loss
@@ -49,5 +55,3 @@ class QHead(Head):
             "Dense (num outputs = {})".format(self.num_actions)
         ]
         return '\n'.join(result)
-
-