diff --git a/src/helper/utils.rs b/src/helper/utils.rs
index 1995951a..a67dc682 100644
--- a/src/helper/utils.rs
+++ b/src/helper/utils.rs
@@ -16,9 +16,19 @@ pub(crate) fn md5_hash(input: impl AsRef<[u8]>) -> String {
 
 pub(crate) fn unescape_xml_text(e: &quick_xml::events::BytesText<'_>) -> String {
     let decoded = e.decode().unwrap();
-    quick_xml::escape::unescape(decoded.as_ref())
-        .unwrap()
-        .into_owned()
+    let mut remaining = decoded.as_ref();
+    let mut output = String::with_capacity(remaining.len());
+    // read_text_into retains XML markup. CDATA contains literal characters,
+    // including ampersands that must neither be decoded nor rejected.
+    while let Some(start) = remaining.find("<![CDATA[") {
+        output.push_str(&quick_xml::escape::unescape(&remaining[..start]).unwrap());
+        let content = &remaining[start + 9..];
+        let end = content.find("]]>").expect("unterminated CDATA section");
+        output.push_str(&content[..end]);
+        remaining = &content[end + 3..];
+    }
+    output.push_str(&quick_xml::escape::unescape(remaining).unwrap());
+    output
 }
 
 /// A macro that implements the `From` trait for converting from one error type
diff --git a/tests/cdata_text.rs b/tests/cdata_text.rs
new file mode 100644
index 00000000..b7f00055
--- /dev/null
+++ b/tests/cdata_text.rs
@@ -0,0 +1,79 @@
+use std::io::{
+    Cursor,
+    Read,
+    Write,
+};
+
+use umya_spreadsheet::{
+    reader,
+    writer,
+};
+use zip::{
+    ZipArchive,
+    ZipWriter,
+    write::SimpleFileOptions,
+};
+
+fn workbook_with_header_text(content: &str) -> Vec<u8> {
+    let mut book = umya_spreadsheet::new_file();
+    let sheet = book.sheet_mut(0).unwrap();
+    sheet.cell_mut("A1").set_value("CDATA regression");
+    sheet
+        .header_footer_mut()
+        .odd_header_mut()
+        .set_value("HEADER_PLACEHOLDER");
+    sheet
+        .header_footer_mut()
+        .odd_footer_mut()
+        .set_value("FOOTER_PLACEHOLDER");
+    let mut original = Vec::new();
+    writer::xlsx::write_writer(&book, &mut original).unwrap();
+    let mut source = ZipArchive::new(Cursor::new(original)).unwrap();
+    let mut output = ZipWriter::new(Cursor::new(Vec::new()));
+    for index in 0..source.len() {
+        let mut part = source.by_index(index).unwrap();
+        let mut bytes = Vec::new();
+        part.read_to_end(&mut bytes).unwrap();
+        if part.name() == "xl/worksheets/sheet1.xml" {
+            bytes = String::from_utf8(bytes)
+                .unwrap()
+                .replace("HEADER_PLACEHOLDER", content)
+                .replace("FOOTER_PLACEHOLDER", content)
+                .into_bytes();
+        }
+        output
+            .start_file(part.name(), SimpleFileOptions::default())
+            .unwrap();
+        output.write_all(&bytes).unwrap();
+    }
+    output.finish().unwrap().into_inner()
+}
+
+#[test]
+fn header_footer_text_preserves_cdata_and_decodes_only_outside_entities() {
+    for (xml, expected) in [
+        (
+            "&amp;CResearch &amp; Development",
+            "&CResearch & Development",
+        ),
+        (
+            "<![CDATA[&C&BResearch & Development&B]]>",
+            "&C&BResearch & Development&B",
+        ),
+        (
+            "&amp;LBefore <![CDATA[&amp; literal <tag>]]> &#65;<![CDATA[&Rafter]]>",
+            "&LBefore &amp; literal <tag> A&Rafter",
+        ),
+        (
+            "<![CDATA[]]>&amp;C<![CDATA[One]]><![CDATA[Two]]>",
+            "&COneTwo",
+        ),
+        ("&lt;![CDATA[ordinary]]&gt;", "<![CDATA[ordinary]]>"),
+    ] {
+        let bytes = workbook_with_header_text(xml);
+        let book = reader::xlsx::read_reader(Cursor::new(bytes), true).unwrap();
+        let header_footer = book.sheet(0).unwrap().header_footer();
+        assert_eq!(header_footer.odd_header().value(), expected, "{xml}");
+        assert_eq!(header_footer.odd_footer().value(), expected, "{xml}");
+    }
+}