diff --git a/examples/ale/train_dqn_ale.py b/examples/ale/train_dqn_ale.py
index a264542fb..56d30efcc 100644
--- a/examples/ale/train_dqn_ale.py
+++ b/examples/ale/train_dqn_ale.py
@@ -78,31 +78,45 @@ def parse_agent(agent):
 
 def main():
     parser = argparse.ArgumentParser()
-    parser.add_argument('--env', type=str, default='BreakoutNoFrameskip-v4')
+    parser.add_argument('--env', type=str, default='BreakoutNoFrameskip-v4',
+                        help='OpenAI Atari domain to perform algorithm on.')
     parser.add_argument('--outdir', type=str, default='results',
                         help='Directory path to save output files.'
                              ' If it does not exist, it will be created.')
     parser.add_argument('--seed', type=int, default=0,
                         help='Random seed [0, 2 ** 31)')
-    parser.add_argument('--gpu', type=int, default=0)
+    parser.add_argument('--gpu', type=int, default=0,
+                        help='GPU to use, set to -1 if no GPU.')
     parser.add_argument('--demo', action='store_true', default=False)
     parser.add_argument('--load', type=str, default=None)
     parser.add_argument('--final-exploration-frames',
-                        type=int, default=10 ** 6)
-    parser.add_argument('--final-epsilon', type=float, default=0.01)
-    parser.add_argument('--eval-epsilon', type=float, default=0.001)
+                        type=int, default=10 ** 6,
+                        help='Timesteps after which we stop ' +
+                        'annealing exploration rate')
+    parser.add_argument('--final-epsilon', type=float, default=0.01,
+                        help='Final value of epsilon during training.')
+    parser.add_argument('--eval-epsilon', type=float, default=0.001,
+                        help='Exploration epsilon used during eval episodes.')
     parser.add_argument('--noisy-net-sigma', type=float, default=None)
     parser.add_argument('--arch', type=str, default='doubledqn',
-                        choices=['nature', 'nips', 'dueling', 'doubledqn'])
-    parser.add_argument('--steps', type=int, default=5 * 10 ** 7)
+                        choices=['nature', 'nips', 'dueling', 'doubledqn'],
+                        help='Network architecture to use.')
+    parser.add_argument('--steps', type=int, default=5 * 10 ** 7,
+                        help='Total number of timesteps to train the agent.')
     parser.add_argument('--max-episode-len', type=int,
                         default=30 * 60 * 60 // 4,  # 30 minutes with 60/4 fps
-                        help='Maximum number of steps for each episode.')
-    parser.add_argument('--replay-start-size', type=int, default=5 * 10 ** 4)
+                        help='Maximum number of timesteps for each episode.')
+    parser.add_argument('--replay-start-size', type=int, default=5 * 10 ** 4,
+                        help='Minimum replay buffer size before ' +
+                        'performing gradient updates.')
     parser.add_argument('--target-update-interval',
-                        type=int, default=3 * 10 ** 4)
-    parser.add_argument('--eval-interval', type=int, default=10 ** 5)
-    parser.add_argument('--update-interval', type=int, default=4)
+                        type=int, default=3 * 10 ** 4,
+                        help='Frequency (in timesteps) at which ' +
+                        'the target network is updated.')
+    parser.add_argument('--eval-interval', type=int, default=10 ** 5,
+                        help='Frequency (in timesteps) of evaluation phase.')
+    parser.add_argument('--update-interval', type=int, default=4,
+                        help='Frequency (in timesteps) of network updates.')
     parser.add_argument('--eval-n-runs', type=int, default=10)
     parser.add_argument('--no-clip-delta',
                         dest='clip_delta', action='store_false')
@@ -117,7 +131,7 @@ def main():
                         help='Monitor env. Videos and additional information'
                              ' are saved as output files.')
     parser.add_argument('--lr', type=float, default=2.5e-4,
-                        help='Learning rate')
+                        help='Learning rate.')
     parser.add_argument('--prioritized', action='store_true', default=False,
                         help='Use prioritized experience replay.')
     args = parser.parse_args()