diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py
index b208efc080..110bea5acd 100644
--- a/pytorch_lightning/models/trainer.py
+++ b/pytorch_lightning/models/trainer.py
@@ -164,6 +164,9 @@ class Trainer(TrainerIO):
         model.zero_grad()
         model.eval()
 
+        # set the model step
+        model.step_split = 'val'
+
         # disable gradients to save memory
         torch.set_grad_enabled(False)
 
@@ -183,7 +186,7 @@ class Trainer(TrainerIO):
             # -----------------
             # RUN VALIDATION STEP
             # -----------------
-            output = model.validation_step(data_batch, batch_i)
+            output = model(data_batch, batch_i)
             outputs.append(output)
 
             # batch done
diff --git a/pytorch_lightning/root_module/root_module.py b/pytorch_lightning/root_module/root_module.py
index 467aeeadf5..b9a6cb4d8e 100644
--- a/pytorch_lightning/root_module/root_module.py
+++ b/pytorch_lightning/root_module/root_module.py
@@ -25,6 +25,7 @@ class RootModule(GradInformation, ModelIO, OptimizerConfig, ModelHooks):
         self.gradient_clip = hparams.gradient_clip
         self.num = 2
         self.trainer = None
+        self.step_split = 'train'
 
         # track if gpu was requested for checkpointing
         self.on_gpu = False
@@ -49,7 +50,14 @@ class RootModule(GradInformation, ModelIO, OptimizerConfig, ModelHooks):
         :param x:
         :return:
         """
-        raise NotImplementedError
+
+        # route the forward call to the correct step type
+        if self.step_split == 'train':
+            return self.training_step(*args, **kwargs)
+        elif self.step_split == 'val':
+            return self.validation_step(*args, **kwargs)
+        else:
+            raise NotImplementedError
 
     def validation_step(self, data_batch, batch_nb):
         """