diff --git a/dask/base.py b/dask/base.py
index c144f367d3a..72426cfb931 100644
--- a/dask/base.py
+++ b/dask/base.py
@@ -1141,8 +1141,7 @@ def normalize_series(s):
 
     @normalize_token.register(pd.DataFrame)
     def normalize_dataframe(df):
-        mgr = df._data
-
+        mgr = df._mgr
         data = list(mgr.arrays) + [df.columns, df.index]
         return list(map(normalize_token, data))