diff --git a/fuzz/.gitignore b/fuzz/.gitignore
new file mode 100644
index 0000000..1a45eee
--- /dev/null
+++ b/fuzz/.gitignore
@@ -0,0 +1,4 @@
+target
+corpus
+artifacts
+coverage
diff --git a/fuzz/Cargo.toml b/fuzz/Cargo.toml
new file mode 100644
index 0000000..1acc864
--- /dev/null
+++ b/fuzz/Cargo.toml
@@ -0,0 +1,22 @@
+[package]
+name = "minifier-fuzz"
+version = "0.0.0"
+publish = false
+edition = "2021"
+
+[package.metadata]
+cargo-fuzz = true
+
+[dependencies]
+libfuzzer-sys = "0.4"
+cssparser = "0.34.0"
+
+[dependencies.minifier]
+path = ".."
+
+[[bin]]
+name = "cssparser"
+path = "fuzz_targets/cssparser.rs"
+test = false
+doc = false
+bench = false
diff --git a/fuzz/fuzz_targets/cssparser.rs b/fuzz/fuzz_targets/cssparser.rs
new file mode 100644
index 0000000..1f176fc
--- /dev/null
+++ b/fuzz/fuzz_targets/cssparser.rs
@@ -0,0 +1,86 @@
+// This file is almost entirely lifted from
+// https://github.com/servo/rust-cssparser/blob/fa6f5eb23f058c6fce444ac781b0b380003fdb59/fuzz/fuzz_targets/cssparser.rs
+// but we compare minified and unminified code instead.
+//
+// This parser branches on ascii characters, so you'll find interesting corner cases more quickly with
+// cargo fuzz run cssparser -- -only_ascii=1
+#![no_main]
+
+use cssparser::*;
+
+const DEBUG: bool = false;
+
+fn parse_and_serialize(input: &str) -> String {
+    let mut input = ParserInput::new(input);
+    let mut parser = Parser::new(&mut input);
+    let mut serialization = String::new();
+    let result = do_parse_and_serialize(
+        &mut parser,
+        TokenSerializationType::nothing(),
+        &mut serialization,
+        0,
+    );
+    if result.is_err() {
+        return String::new();
+    }
+    serialization
+}
+
+fn do_parse_and_serialize<'i>(
+    input: &mut Parser<'i, '_>,
+    mut previous_token_type: TokenSerializationType,
+    serialization: &mut String,
+    indent_level: usize,
+) -> Result<(), ParseError<'i, ()>> {
+    loop {
+        let token = input.next();
+        let token = match token {
+            Ok(token) => token,
+            Err(..) => break,
+        };
+        if DEBUG {
+            for _ in 0..indent_level {
+                print!(" ");
+            }
+            println!("{:?}", token);
+        }
+        if token.is_parse_error() {
+            let token = token.clone();
+            return Err(input.new_unexpected_token_error(token))
+        }
+        let token_type = token.serialization_type();
+        if previous_token_type.needs_separator_when_before(token_type) {
+            serialization.push_str("/**/");
+        }
+        previous_token_type = token_type;
+        token.to_css(serialization).unwrap();
+        let closing_token = match token {
+            Token::Function(_) | Token::ParenthesisBlock => Token::CloseParenthesis,
+            Token::SquareBracketBlock => Token::CloseSquareBracket,
+            Token::CurlyBracketBlock => Token::CloseCurlyBracket,
+            _ => continue,
+        };
+
+        input.parse_nested_block(|input| -> Result<_, ParseError<()>> {
+            do_parse_and_serialize(input, previous_token_type, serialization, indent_level + 1)
+        })?;
+
+        closing_token.to_css(serialization).unwrap();
+    }
+    Ok(())
+}
+
+fn fuzz(data: &str) {
+    let unminified = parse_and_serialize(data);
+    if unminified != "" {
+        let Ok(minified) = minifier::css::minify(data) else { return };
+        let minified = minified.to_string();
+        eprintln!("{minified:?}");
+        let minified = parse_and_serialize(&minified);
+        assert_eq!(unminified, minified);
+    }
+}
+
+libfuzzer_sys::fuzz_target!(|data: &str| {
+    fuzz(data);
+});
\ No newline at end of file