diff --git a/README.md b/README.md
index 376305a..870a3e7 100644
--- a/README.md
+++ b/README.md
@@ -285,6 +285,7 @@ It's important to have a bridge from decord to popular deep learning frameworks
 -   Apache MXNet (Done)
 -   Pytorch (Done)
 -   TensorFlow (Done)
+-   Paddle (Done)
 
 Using bridges for deep learning frameworks are simple, for example, one can set the default tensor output to `mxnet.ndarray`:
 
@@ -297,9 +298,10 @@ print('native output:', type(vr[0]), vr[0].shape)
 decord.bridge.set_bridge('mxnet')
 print(type(vr[0], vr[0].shape))
 # <class 'mxnet.ndarray.ndarray.NDArray'> (240, 426, 3)
-# or pytorch and tensorflow(>=2.2.0)
+# or pytorch, tensorflow(>=2.2.0) and paddle
 decord.bridge.set_bridge('torch')
 decord.bridge.set_bridge('tensorflow')
+decord.bridge.set_bridge('paddle')
 # or back to decord native format
 decord.bridge.set_bridge('native')
 ```
diff --git a/python/decord/bridge/__init__.py b/python/decord/bridge/__init__.py
index 7684bb1..36b9cbf 100644
--- a/python/decord/bridge/__init__.py
+++ b/python/decord/bridge/__init__.py
@@ -6,6 +6,7 @@
 from .torchdl import to_torch, from_torch
 from .tf import to_tensorflow, from_tensorflow
 from .tvm import to_tvm, from_tvm
+from .paddle import to_paddle, from_paddle
 
 _BRIDGE_TYPES = {
     'native': (lambda x: x, lambda x: x),
@@ -13,6 +14,7 @@
     'torch': (to_torch, from_torch),
     'tensorflow': (to_tensorflow, from_tensorflow),
     'tvm': (to_tvm, from_tvm),
+    'paddle': (to_paddle, from_paddle),
 }
 
 _CURRENT_BRIDGE = threading.local()
@@ -74,3 +76,6 @@ def use_tensorflow():
 
 def use_tvm():
     return _BridgeScope('tvm')
+
+def use_paddle():
+    return _BridgeScope('paddle')
diff --git a/python/decord/bridge/paddle.py b/python/decord/bridge/paddle.py
new file mode 100644
index 0000000..e6c81c6
--- /dev/null
+++ b/python/decord/bridge/paddle.py
@@ -0,0 +1,34 @@
+"""DECORD Paddle bridge"""
+from __future__ import absolute_import
+
+from .._ffi._ctypes.ndarray import _from_dlpack
+
+def try_import_paddle():
+    """Try import paddle at runtime.
+
+    Returns
+    -------
+    paddle module if found. Raise ImportError otherwise
+    """
+    msg = "paddle is required, for installation guide, please checkout:\n \
+        https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html"
+    try:
+        return __import__('paddle.utils.dlpack', fromlist=['object'])
+    except ImportError as e:
+        if not message:
+            raise e
+        raise ImportError(message)
+
+def to_paddle(decord_arr):
+    """From decord to paddle.
+    The tensor will share the memory with the object represented in the dlpack.
+    Note that each dlpack can only be consumed once."""
+    dlpack = try_import_paddle()
+    return dlpack.from_dlpack(decord_arr.to_dlpack())
+
+def from_paddle(tensor):
+    """From paddle to decord.
+    The dlpack shares the tensors memory.
+    Note that each dlpack can only be consumed once."""
+    dlpack = try_import_paddle()
+    return _from_dlpack(dlpack.to_dlpack(tensor))
\ No newline at end of file
diff --git a/tests/python/unittests/test_bridges.py b/tests/python/unittests/test_bridges.py
index bb18536..99bf2be 100644
--- a/tests/python/unittests/test_bridges.py
+++ b/tests/python/unittests/test_bridges.py
@@ -59,6 +59,19 @@ def test_tvm_bridge():
     except ImportError:
         print('Skip test tvm bridge as tvm is not found')
 
+def test_paddle_bridge():
+    try:
+        from decord.bridge.paddle import try_import_paddle
+        import paddle
+        vr = _get_default_test_video()
+        with use_paddle():
+            frame = vr[0]
+            assert isinstance(frame, paddle.Tensor), type(frame)
+            native_frame = bridge_in(frame)
+        assert isinstance(native_frame, decord.nd.NDArray), type(native_frame)
+    except ImportError:
+        print('Skip test paddle bridge as paddle is not found')
+
 def test_threaded_bridge():
     # issue #85
     from decord import cpu, gpu