diff --git a/neuralmonkey/runners/tensor_runner.py b/neuralmonkey/runners/tensor_runner.py
index 21a4470dd..7ef7ec2ce 100644
--- a/neuralmonkey/runners/tensor_runner.py
+++ b/neuralmonkey/runners/tensor_runner.py
@@ -17,11 +17,13 @@ def __init__(self,
                  all_coders: Set[ModelPart],
                  fetches: FeedDict,
                  batch_dims: Dict[str, int],
-                 select_session: Optional[int]) -> None:
+                 select_session: Optional[int],
+                 single_tensor: bool) -> None:
         self._all_coders = all_coders
         self._fetches = fetches
         self._batch_dims = batch_dims
         self._select_session = select_session
+        self._single_tensor = single_tensor
 
         self.result = None  # type: Optional[ExecutionResult]
 
@@ -70,6 +72,10 @@ def _fetch_values_from_session(self, sess_results: Dict) -> List:
         batched = [dict(zip(transposed, col))
                    for col in zip(*transposed.values())]
 
+        if self._single_tensor:
+            # extract the only item from each dict
+            batched = [next(iter(d.values())) for d in batched]
+
         return batched
 
 
@@ -90,7 +96,8 @@ def __init__(self,
                  tensors_by_ref: List[tf.Tensor],
                  batch_dims_by_name: List[int],
                  batch_dims_by_ref: List[int],
-                 select_session: int = None) -> None:
+                 select_session: int = None,
+                 single_tensor: bool = False) -> None:
         """Construct a new ``TensorRunner`` object.
 
         Note that at this time, one must specify the toplevel objects so that
@@ -120,15 +127,24 @@ def __init__(self,
                 in case of ensembling. When not used, tensors from all sessions
                 are stored. In case of a single session, this option has no
                 effect.
+            single_tensor: If `True`, it is assumed that only one tensor is to
+                be fetched, and the execution result will consist of this
+                tensor only. If `False`, the result will be a dict mapping
+                tensor names to NumPy arrays.
         """
         check_argument_types()
         BaseRunner[ModelPart].__init__(self, output_series, toplevel_modelpart)
 
+        total_tensors = len(tensors_by_name) + len(tensors_by_ref)
+        if single_tensor and total_tensors > 1:
+            raise ValueError("single_tensor is True, but {} tensors were given".format(total_tensors))
+
         self._names = tensors_by_name
         self._tensors = tensors_by_ref
         self._batch_dims_name = batch_dims_by_name
         self._batch_dims_ref = batch_dims_by_ref
         self._select_session = select_session
+        self._single_tensor = single_tensor
 
         log("Blessing toplevel tensors for tensor runner:")
         for tensor in toplevel_tensors:
@@ -159,7 +175,7 @@ def get_executable(self,
 
         return TensorExecutable(
             self.all_coders, self._fetches, self._batch_ids,
-            self._select_session)
+            self._select_session, self._single_tensor)
     # pylint: enable=unused-argument
 
     @property
@@ -206,4 +222,5 @@ def __init__(self,
             tensors_by_ref=[tensor_to_get],
             batch_dims_by_name=[],
             batch_dims_by_ref=[0],
-            select_session=select_session)
+            select_session=select_session,
+            single_tensor=True)