diff --git a/pandas/io/formats/excel.py b/pandas/io/formats/excel.py
index f2cf1a4838c36..96f55ea6451e8 100644
--- a/pandas/io/formats/excel.py
+++ b/pandas/io/formats/excel.py
@@ -116,8 +116,8 @@ class CSSToExcelConverter:
     focusing on font styling, backgrounds, borders and alignment.
 
     Operates by first computing CSS styles in a fairly generic
-    way (see :meth:`compute_css`) then determining Excel style
-    properties from CSS properties (see :meth:`build_xlstyle`).
+    way (see :meth: `compute_css`) then determining Excel style
+    properties from CSS properties (see :meth: `build_xlstyle`).
 
     Parameters
     ----------
@@ -594,7 +594,11 @@ def _format_value(self, val):
             elif missing.isneginf_scalar(val):
                 val = f"-{self.inf_rep}"
             elif self.float_format is not None:
-                val = float(self.float_format % val)
+                val = self.float_format % val
+            else:
+                # respecter l'affichage par défaut de pandas (console)
+                val = repr(val)
+
         if getattr(val, "tzinfo", None) is not None:
             raise ValueError(
                 "Excel does not support datetimes with "
@@ -616,7 +620,20 @@ def _format_header_mi(self) -> Iterable[ExcelCell]:
 
         columns = self.columns
         merge_columns = self.merge_cells in {True, "columns"}
-        level_strs = columns._format_multi(sparsify=merge_columns, include_names=False)
+
+        # Replace NaN column header values with a non-breaking space so
+        # Excel output matches console display (see user's _fix_headers).
+        NBSP = "\u00A0"
+        if isinstance(columns, MultiIndex):
+            fixed_levels = []
+            for lvl in range(columns.nlevels):
+                vals = columns.get_level_values(lvl)
+                fixed_levels.append([NBSP if missing.isna(v) else str(v) for v in vals])
+            fixed_columns = MultiIndex.from_arrays(fixed_levels, names=columns.names)
+        else:
+            fixed_columns = Index([NBSP if missing.isna(v) else str(v) for v in columns], name=columns.name)
+
+        level_strs = fixed_columns._format_multi(sparsify=merge_columns, include_names=False)
         level_lengths = get_level_lengths(level_strs)
         coloffset = 0
         lnum = 0
@@ -625,17 +642,24 @@ def _format_header_mi(self) -> Iterable[ExcelCell]:
             coloffset = self.df.index.nlevels - 1
 
         for lnum, name in enumerate(columns.names):
+            val = NBSP if missing.isna(name) else str(name)
             yield ExcelCell(
                 row=lnum,
                 col=coloffset,
-                val=name,
+                val=val,
                 style=None,
             )
 
-        for lnum, (spans, levels, level_codes) in enumerate(
-            zip(level_lengths, columns.levels, columns.codes)
+
+
+        # Iterate the fixed_columns levels/codes so values already have
+        # NaNs replaced by NBSP (and are strings).
+        for lnum, (spans, level, codes) in enumerate(
+            zip(level_lengths, fixed_columns.levels, fixed_columns.codes)
         ):
-            values = levels.take(level_codes)
+            # level.take(codes) on fixed_columns.levels yields string values
+            values = level.take(codes).to_numpy()
+
             for i, span_val in spans.items():
                 mergestart, mergeend = None, None
                 if merge_columns and span_val > 1:
@@ -652,6 +676,7 @@ def _format_header_mi(self) -> Iterable[ExcelCell]:
                     mergestart=mergestart,
                     mergeend=mergeend,
                 )
+
         self.rowcounter = lnum
 
     def _format_header_regular(self) -> Iterable[ExcelCell]:
@@ -673,6 +698,12 @@ def _format_header_regular(self) -> Iterable[ExcelCell]:
                     )
                 colnames = self.header
 
+            # Normalize NaN column labels to a non-breaking space so Excel
+            # header output matches console display (same behavior as
+            # applied to MultiIndex headers in _format_header_mi).
+            NBSP = "\u00A0"
+            colnames = [NBSP if missing.isna(v) else str(v) for v in colnames]
+
             for colindex, colname in enumerate(colnames):
                 yield CssExcelCell(
                     row=self.rowcounter,