diff --git a/bluefog/torch/mpi_ops.py b/bluefog/torch/mpi_ops.py
index 4c6b1dbc..2b50d642 100644
--- a/bluefog/torch/mpi_ops.py
+++ b/bluefog/torch/mpi_ops.py
@@ -21,6 +21,7 @@
 import torch
 
 from bluefog.torch import mpi_lib  # C library
+from bluefog.torch.utility import deprecated_function_arg
 from bluefog.common.basics import BlueFogBasics, logger
 from bluefog.common.topology_util import GetRecvWeights
 
@@ -528,7 +529,8 @@ def _neighbor_allreduce_nonblocking(tensor, output, self_weight, src_weights,
     _handle_map[handle] = (tensor, output)
     return handle
 
-
+@deprecated_function_arg(arg_name="neighbor_weights", fix="Use src_weights instead")
+@deprecated_function_arg(arg_name="send_neighbors", fix="Use dst_weights instead")
 def neighbor_allreduce(tensor: torch.Tensor, *,
                        self_weight: Optional[float] = None,
                        src_weights: Optional[Dict[int, float]] = None,
@@ -548,9 +550,9 @@ def neighbor_allreduce(tensor: torch.Tensor, *,
 
     Arguments:
         tensor: A tensor to execute weighted average with neighbors.
-        self_weight: The weight for self node, used with neighbor_weights.
+        self_weight: The weight for self node, used with src_weights.
         src_weights: The weights for in-neighbor nodes, used with self weight.
-            If neighbor_weights is presented, the return tensor will return the weighted average
+            If src_weights is presented, the return tensor will return the weighted average
             defined by these weights and the self_weight. If not, the return tensor will return
             the weighted average defined by the topology weights is provided or uniformly average.
             The data structure of weights should be {rank : weight} and rank has to belong to the
@@ -560,7 +562,7 @@ def neighbor_allreduce(tensor: torch.Tensor, *,
             part of (out-)neighbors will be sent to. If set to be a list, assume all the weights
             are one. In this mode, this node sends its value to partial neighbors listed in this
             variable in a dynamic graph, and `self_weight` and `src_weights` must be present.
-        enable_topo_check: When send_neighbors is present, enabling this option checks if the
+        enable_topo_check: When dst_weights is present, enabling this option checks if the
             sending and recieving neighbors match with each other. Disabling this check can boost
             the performance.
         name: A name of the reduction operation.
@@ -584,6 +586,8 @@ def neighbor_allreduce(tensor: torch.Tensor, *,
     return synchronize(handle)
 
 
+@deprecated_function_arg(arg_name="neighbor_weights", fix="Use src_weights instead")
+@deprecated_function_arg(arg_name="send_neighbors", fix="Use dst_weights instead")
 def neighbor_allreduce_nonblocking(tensor: torch.Tensor, *,
                                    self_weight: Optional[float] = None,
                                    src_weights: Optional[Dict[int, float]] = None,
@@ -603,9 +607,9 @@ def neighbor_allreduce_nonblocking(tensor: torch.Tensor, *,
 
     Arguments:
         tensor: A tensor to execute weighted average with neighbors.
-        self_weight: The weight for self node, used with neighbor_weights.
+        self_weight: The weight for self node, used with src_weights.
         src_weights: The weights for in-neighbor nodes, used with self weight.
-            If neighbor_weights is presented, the return tensor will return the weighted average
+            If src_weights is presented, the return tensor will return the weighted average
             defined by these weights and the self_weight. If not, the return tensor will return
             the weighted average defined by the topology weights is provided or uniformly average.
             The data structure of weights should be {rank : weight} and rank has to belong to the
@@ -615,7 +619,7 @@ def neighbor_allreduce_nonblocking(tensor: torch.Tensor, *,
             part of (out-)neighbors will be sent to. If set to be a list, assume all the weights
             are one. In this mode, this node sends its value to partial neighbors listed in this
             variable in a dynamic graph, and `self_weight` and `src_weights` must be present.
-        enable_topo_check: When send_neighbors is present, enabling this option checks if the
+        enable_topo_check: When dst_weights is present, enabling this option checks if the
             sending and recieving neighbors match with each other. Disabling this check can boost
             the performance.
         name: A name of the neighbor_allreduce operation.
diff --git a/bluefog/torch/utility.py b/bluefog/torch/utility.py
index 73337207..7d35f56a 100644
--- a/bluefog/torch/utility.py
+++ b/bluefog/torch/utility.py
@@ -15,12 +15,14 @@
 # ==============================================================================
 
 from typing import Any, List, Optional
+from functools import wraps
 import collections
 
 import numpy as np
 import torch
 import bluefog.torch as bf
 
+
 def broadcast_parameters(params, root_rank):
     """
     Broadcasts the parameters from root rank to all other processes.
@@ -212,3 +214,16 @@ def _from_tensor():
     for key, p in params:
         if key in callbacks:
             callbacks[key]()
+
+
+def deprecated_function_arg(arg_name: str, fix: str):
+    def deprecated_decorator(f):
+        @wraps(f)
+        def wrapper(*arg, **kwargs):
+            if arg_name in kwargs:
+                raise TypeError(f"{arg_name} is deprecated in {f.__name__}: {fix}")
+            return f(*arg, **kwargs)
+
+        return wrapper
+
+    return deprecated_decorator