diff --git a/protocol/pandas_implementation.py b/protocol/pandas_implementation.py
index a016d25c..16fdf432 100644
--- a/protocol/pandas_implementation.py
+++ b/protocol/pandas_implementation.py
@@ -435,10 +435,32 @@ def _dtype_from_pandasdtype(self, dtype) -> Tuple[enum.IntEnum, int, str, str]:
             raise NotImplementedError(f"Data type {dtype} not handled yet")
 
         bitwidth = dtype.itemsize * 8
-        format_str = dtype.str
+        format_str = self._format_str(dtype.str)
         endianness = dtype.byteorder if not kind == _k.CATEGORICAL else '='
         return (kind, bitwidth, format_str, endianness)
 
+    def _format_str(self, format_str) -> str:
+        """
+        Mapping of NumPy format strings to
+        Apache Arrow C Data Interface format strings.
+        'O' categorical mapped as 'U': large utf-8 string for now
+        """
+        _ints = {8: 'c', 16: 's', 32: 'i', 64: 'l'}
+        _uints = {8: 'C', 16: 'S', 32: 'I', 64: 'L'}
+        _floats = {16: 'e', 32: 'f', 64: 'g'}
+        _np_dtypes = {'i': _ints, 'u': _uints, 'f': _floats, 'b': {8: 'b'}, 'O': {64: 'U'}} 
+        
+        dt = np.dtype(format_str)
+        if dt.byteorder == '>':
+            raise ValueError(f"Big-endian not supported by exchange"
+                                 "protocol")
+            
+        arrow_format_str = _np_dtypes.get(dt.kind, {}).get(dt.itemsize*8)
+        
+        if arrow_format_str is None:
+            raise NotImplementedError(f"Format string {format_str} not handled yet")
+            
+        return arrow_format_str
 
     @property
     def describe_categorical(self) -> Dict[str, Any]:
@@ -880,6 +902,16 @@ def test_metadata():
     assert_dataframe_equal(df.__dataframe__(), df)
     tm.assert_frame_equal(df, df2)
 
+def test_fromat_str():
+    df = pd.DataFrame(data=dict(a=[1, 2, 3], B=[3, 4, 5],
+                                c=[1.5, 2.5, 3.5], D=["a", "b", "cdef"]))
+    df["B"] = df["B"].astype("category")
+    df["D"] = df["D"].astype("object")
+    
+    format_strings = {'a': 'l', 'B': 'U', 'c': 'g', 'D': 'u'}
+    for col in df.columns.tolist():
+        column = _PandasColumn(df[col])
+        assert column.dtype[2] == format_strings[col]
 
 if __name__ == '__main__':
     test_categorical_dtype()