diff --git a/RELEASE.md b/RELEASE.md
index 17434a840b4..f2d3c3c6efe 100644
--- a/RELEASE.md
+++ b/RELEASE.md
@@ -11,7 +11,8 @@
   [TensorFloat-32](https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/).
   Specifically, inputs to such ops are rounded from 23 bits of precision to 10
   bits of precision. This is unlikely to cause issues in practice for deep
-  learning models. TensorFloat-32 can be disabled by running
+  learning models. In some cases, TensorFloat-32 is also used for complex64 ops.
+  TensorFloat-32 can be disabled by running
   `config.experimental.enable_tensor_float_32_execution(False)`. The "Major
   Features and Improvements" section has more details.
 * The byte layout for string tensors across the C-API has been updated to match
diff --git a/tensorflow/python/framework/config.py b/tensorflow/python/framework/config.py
index 4ad090eaf96..2691665ffce 100644
--- a/tensorflow/python/framework/config.py
+++ b/tensorflow/python/framework/config.py
@@ -81,6 +81,9 @@ def enable_tensor_float_32_execution(enabled):
   be added in the future. As a result, precision of float32 ops may decrease in
   minor versions of TensorFlow.
 
+  TensorFloat-32 is also used for some complex64 ops. Currently, TensorFloat-32
+  is used in fewer cases for complex64 as it is for float32.
+
   Args:
     enabled: Bool indicating whether to enable TensorFloat-32 execution.
   """