diff --git a/src/tools/compiletest/src/header.rs b/src/tools/compiletest/src/header.rs
index 67ecdaf9e5e13..48149e3b897e1 100644
--- a/src/tools/compiletest/src/header.rs
+++ b/src/tools/compiletest/src/header.rs
@@ -978,10 +978,10 @@ impl Config {
         }
     }
 
-    fn parse_custom_normalization(&self, line: &str) -> Option<NormalizeRule> {
+    fn parse_custom_normalization(&self, raw_directive: &str) -> Option<NormalizeRule> {
         // FIXME(Zalathar): Integrate name/value splitting into `DirectiveLine`
         // instead of doing it here.
-        let (directive_name, _value) = line.split_once(':')?;
+        let (directive_name, raw_value) = raw_directive.split_once(':')?;
 
         let kind = match directive_name {
             "normalize-stdout" => NormalizeKind::Stdout,
@@ -991,11 +991,9 @@ impl Config {
             _ => return None,
         };
 
-        // FIXME(Zalathar): The normalize rule parser should only care about
-        // the value part, not the "line" (which isn't even the whole line).
-        let Some((regex, replacement)) = parse_normalize_rule(line) else {
+        let Some((regex, replacement)) = parse_normalize_rule(raw_value) else {
             panic!(
-                "couldn't parse custom normalization rule: `{line}`\n\
+                "couldn't parse custom normalization rule: `{raw_directive}`\n\
                 help: expected syntax is: `{directive_name}: \"REGEX\" -> \"REPLACEMENT\"`"
             );
         };
@@ -1141,24 +1139,26 @@ enum NormalizeKind {
 /// Parses the regex and replacement values of a `//@ normalize-*` header,
 /// in the format:
 /// ```text
-/// normalize-*: "REGEX" -> "REPLACEMENT"
+/// "REGEX" -> "REPLACEMENT"
 /// ```
-fn parse_normalize_rule(header: &str) -> Option<(String, String)> {
+fn parse_normalize_rule(raw_value: &str) -> Option<(String, String)> {
     // FIXME: Support escaped double-quotes in strings.
     let captures = static_regex!(
         r#"(?x) # (verbose mode regex)
         ^
-        [^:\s]+:\s*             # (header name followed by colon)
+        \s*                     # (leading whitespace)
         "(?<regex>[^"]*)"       # "REGEX"
         \s+->\s+                # ->
         "(?<replacement>[^"]*)" # "REPLACEMENT"
         $
         "#
     )
-    .captures(header)?;
+    .captures(raw_value)?;
     let regex = captures["regex"].to_owned();
     let replacement = captures["replacement"].to_owned();
-    // FIXME: Support escaped new-line in strings.
+    // A `\n` sequence in the replacement becomes an actual newline.
+    // FIXME: Do unescaping in a less ad-hoc way, and perhaps support escaped
+    // backslashes and double-quotes.
     let replacement = replacement.replace("\\n", "\n");
     Some((regex, replacement))
 }
diff --git a/src/tools/compiletest/src/header/tests.rs b/src/tools/compiletest/src/header/tests.rs
index cd7c6f8361ed3..618b66dfd4cb6 100644
--- a/src/tools/compiletest/src/header/tests.rs
+++ b/src/tools/compiletest/src/header/tests.rs
@@ -35,11 +35,14 @@ fn make_test_description<R: Read>(
 
 #[test]
 fn test_parse_normalize_rule() {
-    let good_data = &[(
-        r#"normalize-stderr-32bit: "something (32 bits)" -> "something ($WORD bits)""#,
-        "something (32 bits)",
-        "something ($WORD bits)",
-    )];
+    let good_data = &[
+        (
+            r#""something (32 bits)" -> "something ($WORD bits)""#,
+            "something (32 bits)",
+            "something ($WORD bits)",
+        ),
+        (r#"  " with whitespace"   ->   "    replacement""#, " with whitespace", "    replacement"),
+    ];
 
     for &(input, expected_regex, expected_replacement) in good_data {
         let parsed = parse_normalize_rule(input);
@@ -49,15 +52,15 @@ fn test_parse_normalize_rule() {
     }
 
     let bad_data = &[
-        r#"normalize-stderr-32bit "something (32 bits)" -> "something ($WORD bits)""#,
-        r#"normalize-stderr-16bit: something (16 bits) -> something ($WORD bits)"#,
-        r#"normalize-stderr-32bit: something (32 bits) -> something ($WORD bits)"#,
-        r#"normalize-stderr-32bit: "something (32 bits) -> something ($WORD bits)"#,
-        r#"normalize-stderr-32bit: "something (32 bits)" -> "something ($WORD bits)"#,
-        r#"normalize-stderr-32bit: "something (32 bits)" -> "something ($WORD bits)"."#,
+        r#"something (11 bits) -> something ($WORD bits)"#,
+        r#"something (12 bits) -> something ($WORD bits)"#,
+        r#""something (13 bits) -> something ($WORD bits)"#,
+        r#""something (14 bits)" -> "something ($WORD bits)"#,
+        r#""something (15 bits)" -> "something ($WORD bits)"."#,
     ];
 
     for &input in bad_data {
+        println!("- {input:?}");
         let parsed = parse_normalize_rule(input);
         assert_eq!(parsed, None);
     }
diff --git a/tests/ui/compiletest-self-test/normalize-with-revision.a.run.stderr b/tests/ui/compiletest-self-test/normalize-with-revision.a.run.stderr
new file mode 100644
index 0000000000000..3eb3f6b4e5712
--- /dev/null
+++ b/tests/ui/compiletest-self-test/normalize-with-revision.a.run.stderr
@@ -0,0 +1,2 @@
+1st emitted line
+second emitted line
diff --git a/tests/ui/compiletest-self-test/normalize-with-revision.b.run.stderr b/tests/ui/compiletest-self-test/normalize-with-revision.b.run.stderr
new file mode 100644
index 0000000000000..8d9156480abff
--- /dev/null
+++ b/tests/ui/compiletest-self-test/normalize-with-revision.b.run.stderr
@@ -0,0 +1,2 @@
+first emitted line
+2nd emitted line
diff --git a/tests/ui/compiletest-self-test/normalize-with-revision.rs b/tests/ui/compiletest-self-test/normalize-with-revision.rs
new file mode 100644
index 0000000000000..e1bbbb3eabb70
--- /dev/null
+++ b/tests/ui/compiletest-self-test/normalize-with-revision.rs
@@ -0,0 +1,20 @@
+//! Checks that `[rev] normalize-*` directives affect the specified revision,
+//! and don't affect other revisions.
+//!
+//! This currently relies on the fact that `normalize-*` directives are
+//! applied to run output, not just compiler output. If that ever changes,
+//! this test might need to be adjusted.
+
+//@ edition: 2021
+//@ revisions: a b
+//@ run-pass
+//@ check-run-results
+
+//@ normalize-stderr: "output" -> "emitted"
+//@[a] normalize-stderr: "first" -> "1st"
+//@[b] normalize-stderr: "second" -> "2nd"
+
+fn main() {
+    eprintln!("first output line");
+    eprintln!("second output line");
+}