diff --git a/Changelog.md b/Changelog.md
index 25c9aab1..97fbbc9b 100644
--- a/Changelog.md
+++ b/Changelog.md
@@ -25,6 +25,9 @@
 
 - [#1007]: Remove error-prone implementations of `Hash`, `PartialOrd`, and `Ord` from all
   events except `BytesText` and `BytesCData`, which was introduced in [#1005].
+- [#1017]: Fix panic and wrong output when the indent character is not ASCII.
+  `Indentation` used a character count as a byte index into its indent buffer, and
+  `Serializer::indent` truncated its `char` argument with `as u8`.
 
 ### Misc Changes
 
@@ -35,6 +38,7 @@
 [#1009]: https://github.com/tafia/quick-xml/pull/1009
 [#1010]: https://github.com/tafia/quick-xml/pull/1010
 [#1015]: https://github.com/tafia/quick-xml/pull/1015
+[#1017]: https://github.com/tafia/quick-xml/pull/1017
 
 
 ## 0.42.0 -- 2026-08-22
diff --git a/src/se/mod.rs b/src/se/mod.rs
index 9db3ef40..72c0f538 100644
--- a/src/se/mod.rs
+++ b/src/se/mod.rs
@@ -817,7 +817,7 @@ impl<'w, 'r, W: Write> Serializer<'w, 'r, W> {
     /// );
     /// ```
     pub fn indent(&mut self, indent_char: char, indent_size: usize) -> &mut Self {
-        self.ser.indent = Indent::Owned(Indentation::new(indent_char as u8, indent_size));
+        self.ser.indent = Indent::Owned(Indentation::with_char(indent_char, indent_size));
         self
     }
 
diff --git a/src/writer.rs b/src/writer.rs
index b6a7c32a..9c89edf5 100644
--- a/src/writer.rs
+++ b/src/writer.rs
@@ -678,13 +678,16 @@ pub(crate) struct Indentation {
     indent_size: usize,
     /// Used as a cache for the string used for indentation
     indents: String,
-    /// The current amount of indentation
+    /// The current amount of indentation, in bytes of `indents`
     current_indent_len: usize,
 }
 
 impl Indentation {
     pub fn new(indent_char: u8, indent_size: usize) -> Self {
-        let indent_char = char::from(indent_char);
+        Self::with_char(char::from(indent_char), indent_size)
+    }
+
+    pub fn with_char(indent_char: char, indent_size: usize) -> Self {
         Self {
             should_line_break: false,
             indent_char,
@@ -694,15 +697,21 @@ impl Indentation {
         }
     }
 
+    /// Size of one indentation level in bytes
+    #[inline]
+    const fn level_len(&self) -> usize {
+        self.indent_size * self.indent_char.len_utf8()
+    }
+
     /// Increase indentation by one level
     pub fn grow(&mut self) {
-        self.current_indent_len += self.indent_size;
+        self.current_indent_len += self.level_len();
         self.ensure(self.current_indent_len);
     }
 
     /// Decrease indentation by one level. Do nothing, if level already zero
     pub fn shrink(&mut self) {
-        self.current_indent_len = self.current_indent_len.saturating_sub(self.indent_size);
+        self.current_indent_len = self.current_indent_len.saturating_sub(self.level_len());
     }
 
     /// Returns indent string for current level
@@ -711,8 +720,10 @@ impl Indentation {
     }
 
     /// Returns indent with current indent plus additional indent
+    ///
+    /// `additional_indent` is counted in characters, not bytes.
     pub fn additional(&mut self, additional_indent: usize) -> &str {
-        let new_len = self.current_indent_len + additional_indent;
+        let new_len = self.current_indent_len + additional_indent * self.indent_char.len_utf8();
         self.ensure(new_len);
         &self.indents[..new_len]
     }
diff --git a/tests/writer-indentation.rs b/tests/writer-indentation.rs
index 4e23bbc6..560c79dc 100644
--- a/tests/writer-indentation.rs
+++ b/tests/writer-indentation.rs
@@ -550,3 +550,170 @@ mod in_attributes_multi {
         );
     }
 }
+
+/// `Writer::new_with_indent` converts the indent byte with `char::from`, so bytes
+/// above 0x7F select a character that is two bytes long in UTF-8.
+mod multi_byte_indent_char {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    /// NO-BREAK SPACE is `char::from(0xA0)` and occupies two bytes in UTF-8.
+    const NBSP: char = '\u{a0}';
+
+    #[test]
+    fn nested() {
+        let mut buffer = Vec::new();
+        let mut writer = Writer::new_with_indent(&mut buffer, 0xA0, 2);
+
+        let start = BytesStart::new("paired");
+        let end = start.to_end().into_owned();
+
+        writer
+            .write_event(Event::Start(start.clone()))
+            .expect("write start 1 tag failed");
+        writer
+            .write_event(Event::Start(start))
+            .expect("write start 2 tag failed");
+        writer
+            .write_event(Event::Empty(BytesStart::new("inner")))
+            .expect("write inner tag failed");
+        writer
+            .write_event(Event::End(end.clone()))
+            .expect("write end tag 2 failed");
+        writer
+            .write_event(Event::End(end))
+            .expect("write end tag 1 failed");
+
+        let i = NBSP.to_string().repeat(2);
+        assert_eq!(
+            std::str::from_utf8(&buffer).unwrap(),
+            format!(
+                "<paired>\n{i}<paired>\n{i}{i}<inner/>\n{i}</paired>\n</paired>",
+                i = i
+            )
+        );
+    }
+
+    #[test]
+    fn in_attributes() {
+        let mut buffer = Vec::new();
+        let mut writer = Writer::new_with_indent(&mut buffer, 0xA0, 2);
+
+        writer
+            .create_element("element")
+            .new_line()
+            .with_attribute(("first", "1"))
+            .write_empty()
+            .expect("write tag failed");
+
+        assert_eq!(
+            std::str::from_utf8(&buffer).unwrap(),
+            format!("<element\n{i}first=\"1\"/>", i = NBSP.to_string().repeat(2))
+        );
+    }
+
+    /// An odd `indent_size` makes the byte length of one level an odd number,
+    /// which used to land in the middle of a two byte character and panic.
+    #[test]
+    fn odd_indent_size() {
+        let mut buffer = Vec::new();
+        let mut writer = Writer::new_with_indent(&mut buffer, 0xA0, 1);
+
+        let start = BytesStart::new("paired");
+        writer
+            .write_event(Event::Start(start.clone()))
+            .expect("write start tag failed");
+        writer
+            .write_event(Event::Empty(BytesStart::new("inner")))
+            .expect("write inner tag failed");
+        writer
+            .write_event(Event::End(start.to_end()))
+            .expect("write end tag failed");
+
+        assert_eq!(
+            std::str::from_utf8(&buffer).unwrap(),
+            format!("<paired>\n{i}<inner/>\n</paired>", i = NBSP)
+        );
+    }
+
+    /// Attribute indent is added on top of the current element indent, so the
+    /// conversion must apply to the added level only, not to the sum.
+    #[test]
+    fn in_attributes_nested() {
+        let mut buffer = Vec::new();
+        let mut writer = Writer::new_with_indent(&mut buffer, 0xA0, 2);
+
+        let outer = BytesStart::new("outer");
+        writer
+            .write_event(Event::Start(outer.clone()))
+            .expect("write outer tag failed");
+        writer
+            .create_element("element")
+            .new_line()
+            .with_attribute(("first", "1"))
+            .write_empty()
+            .expect("write tag failed");
+        writer
+            .write_event(Event::End(outer.to_end()))
+            .expect("write end tag failed");
+
+        let i = NBSP.to_string();
+        assert_eq!(
+            std::str::from_utf8(&buffer).unwrap(),
+            format!(
+                "<outer>\n{i2}<element\n{i4}first=\"1\"/>\n</outer>",
+                i2 = i.repeat(2),
+                i4 = i.repeat(4)
+            )
+        );
+    }
+
+    /// ASCII indents must keep working exactly as before.
+    #[test]
+    fn ascii_is_unchanged() {
+        let mut buffer = Vec::new();
+        let mut writer = Writer::new_with_indent(&mut buffer, b' ', 2);
+
+        let start = BytesStart::new("paired");
+        writer
+            .write_event(Event::Start(start.clone()))
+            .expect("write start tag failed");
+        writer
+            .write_event(Event::Empty(BytesStart::new("inner")))
+            .expect("write inner tag failed");
+        writer
+            .write_event(Event::End(start.to_end()))
+            .expect("write end tag failed");
+
+        assert_eq!(
+            std::str::from_utf8(&buffer).unwrap(),
+            "<paired>\n  <inner/>\n</paired>"
+        );
+    }
+
+    /// `Serializer::indent` takes a `char`, so it must not be truncated to a byte.
+    #[cfg(feature = "serialize")]
+    #[test]
+    fn serializer_indent_char_is_not_truncated() {
+        use quick_xml::se::Serializer;
+        use serde::Serialize;
+
+        #[derive(Serialize)]
+        struct Response {
+            message: &'static str,
+        }
+
+        // IDEOGRAPHIC SPACE: three bytes in UTF-8, `as u8` truncates it to NUL.
+        let mut output = String::new();
+        let mut serializer = Serializer::with_root(&mut output, Some("response")).unwrap();
+        serializer.indent('\u{3000}', 2);
+        Response { message: "Success" }
+            .serialize(serializer)
+            .unwrap();
+
+        assert_eq!(
+            output,
+            "<response>\n\u{3000}\u{3000}<message>Success</message>\n</response>"
+        );
+    }
+}