diff --git a/src/garage/tf/algos/npo.py b/src/garage/tf/algos/npo.py
index a02a2df654..4527f4ce51 100644
--- a/src/garage/tf/algos/npo.py
+++ b/src/garage/tf/algos/npo.py
@@ -129,6 +129,9 @@ def __init__(self,
         if pg_loss not in ['vanilla', 'surrogate', 'surrogate_clip']:
             raise ValueError('Invalid pg_loss')
 
+        if self.max_episode_length == None:
+            raise ValueError('max_episode_length must not be None')
+
         self._optimizer = make_optimizer(optimizer, **optimizer_args)
         self._lr_clip_range = float(lr_clip_range)
         self._max_kl_step = float(max_kl_step)
diff --git a/tests/garage/tf/algos/test_npo.py b/tests/garage/tf/algos/test_npo.py
index abeea28de0..b889f22650 100644
--- a/tests/garage/tf/algos/test_npo.py
+++ b/tests/garage/tf/algos/test_npo.py
@@ -115,6 +115,22 @@ def test_npo_with_invalid_no_entropy_configuration(self):
                 entropy_method='no_entropy',
                 policy_ent_coeff=0.02,
             )
+    
+    @pytest.mark.mujoco
+    def test_npo_with_invalid_max_episode_length(self):
+        """Test NPO with invalid max_episode_length."""
+        with pytest.raises(ValueError):
+            env = normalize(
+            GymEnv('InvertedDoublePendulum-v2', max_episode_length=None))
+            NPO(
+                env_spec=env.spec,
+                policy=self.policy,
+                baseline=self.baseline,
+                sampler=self.sampler,
+                discount=0.99,
+                gae_lambda=0.98,
+                policy_ent_coeff=0.0
+            )
 
     def teardown_method(self):
         self.env.close()