diff --git a/fairscale/nn/data_parallel/fully_sharded_data_parallel.py b/fairscale/nn/data_parallel/fully_sharded_data_parallel.py
index e1e294148..1b1b16d30 100644
--- a/fairscale/nn/data_parallel/fully_sharded_data_parallel.py
+++ b/fairscale/nn/data_parallel/fully_sharded_data_parallel.py
@@ -688,7 +688,7 @@ def _cast_buffers(
     @property
     def params_with_grad(self) -> List[Parameter]:
         """[p for p in self.parameters() if p.grad is not None]"""
-        return [p for p in self.parameters() if (p.grad is not None or p.main_grad is not None)]
+        return [p for p in self.parameters() if (p.requires_grad and (p.grad is not None or p.main_grad is not None))]
 
     @torch.no_grad()
     def clip_grad_norm_(