diff --git a/Cargo.toml b/Cargo.toml
index e2874e6..3f79916 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -14,6 +14,7 @@ A Rust library for using Sigsum transparency logs.
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
+base16ct = "0.3.0"
 base64ct = { version = "1.8.0", features = ["alloc"] }
 ed25519-dalek = "2.2.0"
 sha2 = "0.10.8"
diff --git a/src/policy/mod.rs b/src/policy/mod.rs
index 0afd0d9..e27c63b 100644
--- a/src/policy/mod.rs
+++ b/src/policy/mod.rs
@@ -2,6 +2,10 @@ use std::collections::HashMap;
 
 use crate::crypto::{Hash, PublicKey};
 
+mod parsing;
+
+pub use parsing::ParsePolicyError;
+
 /// A Sigsum policy.
 ///
 /// The Sigsum policy dictates if a signed tree head is considered valid (and by extension, if a
diff --git a/src/policy/parsing.rs b/src/policy/parsing.rs
new file mode 100644
index 0000000..91b077b
--- /dev/null
+++ b/src/policy/parsing.rs
@@ -0,0 +1,368 @@
+use std::iter::Enumerate;
+use std::str::Lines;
+
+use super::{Policy, PolicyBuilder};
+use crate::crypto::PublicKey;
+
+#[derive(Debug, thiserror::Error)]
+#[error("line {lineno}: {reason}")]
+pub struct ParsePolicyError {
+    lineno: usize,
+    reason: String,
+}
+
+macro_rules! bail {
+    ($lineno:expr, $($farg:tt)*) => {
+        return Err(ParsePolicyError{lineno: $lineno, reason: format!($($farg)*)})
+    };
+}
+
+type Result<T> = std::result::Result<T, ParsePolicyError>;
+
+impl Policy {
+    /// Parse a policy from its string representation according to the
+    /// [sigsum-go spec](https://git.glasklar.is/sigsum/core/sigsum-go/-/blob/main/doc/policy.md).
+    pub fn parse(data: &str) -> Result<Policy> {
+        parse_policy(data)
+    }
+}
+
+fn parse_policy(data: &str) -> Result<Policy> {
+    let mut builder = PolicyBuilder::new();
+    let lines = PolicyLines::new(data);
+    for (lineno, tokens) in lines {
+        match tokens[..] {
+            ["log", ..] => {
+                let nargs = tokens.len() - 1;
+                if !(1..=2).contains(&nargs) {
+                    bail!(
+                        lineno,
+                        "invalid log rule: expected 1 or 2 arguments, got {nargs}"
+                    );
+                }
+                let pubkey = parse_key(tokens[1], lineno)?;
+                let url = tokens.get(2).map(|s| (*s).into());
+                if let Err(err) = builder.add_log(pubkey, url) {
+                    bail!(lineno, "invalid log rule: {err}")
+                }
+            }
+            ["witness", ..] => {
+                let nargs = tokens.len() - 1;
+                if !(2..=3).contains(&nargs) {
+                    bail!(
+                        lineno,
+                        "invalid witness rule: expected 2 or 3 arguments, got {nargs}"
+                    );
+                }
+                let name = tokens[1];
+                let pubkey = parse_key(tokens[2], lineno)?;
+                let url = tokens.get(3).map(|s| (*s).into());
+                if let Err(err) = builder.add_witness(name.into(), pubkey, url) {
+                    bail!(lineno, "invalid witness rule: {err}")
+                }
+            }
+            ["group", ..] => {
+                let nargs = tokens.len() - 1;
+                if nargs < 3 {
+                    bail!(
+                        lineno,
+                        "invalid group rule: expected at least 3 arguments, got {nargs}"
+                    );
+                }
+                let name = tokens[1];
+                let k = tokens[2];
+                let members: Vec<_> = tokens[3..].iter().map(|n| (*n).into()).collect();
+                let Some(k) = parse_k(k, members.len()) else {
+                    bail!(lineno, "invalid group rule: cannot parse {k} as an integer");
+                };
+                if let Err(err) = builder.add_group(name.into(), k, members) {
+                    bail!(lineno, "invalid group rule: {err}");
+                }
+            }
+            ["quorum", name] => {
+                if name != "none" {
+                    if let Err(err) = builder.set_quorum(name.into()) {
+                        bail!(lineno, "invalid quorum rule: {err}");
+                    }
+                }
+            }
+            ["quorum", ..] => bail!(lineno, "invalid quorum rule: expected exactly one argument"),
+            [unknown, ..] => bail!(lineno, "unknown keyword `{unknown}`"),
+            [] => unreachable!("PolicyLines skips empty lines"),
+        }
+    }
+    Ok(builder.build())
+}
+
+fn parse_k(k: &str, n: usize) -> Option<usize> {
+    if k == "any" {
+        Some(1)
+    } else if k == "all" {
+        Some(n)
+    } else {
+        k.parse().ok()
+    }
+}
+
+fn parse_key(hex: &str, lineno: usize) -> Result<PublicKey> {
+    if base16ct::decoded_len(hex.as_bytes()) != Ok(32) {
+        return Err(ParsePolicyError {
+            lineno,
+            reason: String::from("invalid public key: wrong length"),
+        });
+    }
+    let mut buf = [0; 32];
+    match base16ct::mixed::decode(hex, &mut buf) {
+        Err(_) => Err(ParsePolicyError {
+            lineno,
+            reason: "invalid public key: invalid hex encoding".to_string(),
+        }),
+        Ok(_) => Ok(buf.into()),
+    }
+}
+
+// PolicyLines is essentially a lexer for parsing policy files.
+//
+// It implements a custom iterator that handles most low-level details of parsing policy files,
+// splitting, numbering, filtering and tokenizing lines.
+struct PolicyLines<'a>(Enumerate<Lines<'a>>);
+
+impl<'a> PolicyLines<'a> {
+    fn new(input: &'a str) -> Self {
+        Self(input.lines().enumerate())
+    }
+}
+
+impl<'a> Iterator for PolicyLines<'a> {
+    // Return a pair line number + tokens
+    type Item = (usize, Vec<&'a str>);
+
+    // next() is the only required method
+    fn next(&mut self) -> Option<Self::Item> {
+        loop {
+            let (i, line) = self.0.next()?;
+            let line = line.trim_ascii();
+            if line.is_empty() || line.starts_with('#') {
+                continue;
+            }
+            let tokens = line.split_ascii_whitespace().collect();
+            return Some((i + 1, tokens));
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use hex_literal::hex;
+
+    use super::*;
+
+    #[test]
+    fn policy_lines_empty() {
+        let pls: Vec<(usize, Vec<&str>)> = PolicyLines::new("").collect();
+        assert_eq!(pls, vec![]);
+    }
+
+    #[test]
+    fn policy_lines() {
+        let pls: Vec<(usize, Vec<&str>)> = PolicyLines::new(
+            "# This is a comment\n\
+             foo bar\n\
+             \n\
+             # The next line uses tabs\n\
+             foo\tbar\n\
+             \n\
+             # This one mixes whitespaces\n\
+             foo  \tbar\t  baz \t biz\n\
+             \n\
+             # Heading/trailing spaces\n\
+             \t\n\
+             \t  \n\
+             \t# Indented comment\n\
+             foo\t \n\
+             \n\
+             # Unicode space\n\
+             foo\u{2003}bar\n\
+             \u{2003}\n\
+             ",
+        )
+        .collect();
+        assert_eq!(
+            pls,
+            vec![
+                (2, vec!["foo", "bar"]),
+                (5, vec!["foo", "bar"]),
+                (8, vec!["foo", "bar", "baz", "biz"]),
+                (14, vec!["foo"]),
+                (17, vec!["foo\u{2003}bar"]),
+                (18, vec!["\u{2003}"]),
+            ]
+        );
+    }
+
+    #[test]
+    fn parse_policy_empty() {
+        let expected = PolicyBuilder::new().build();
+        let actual = parse_policy("").unwrap();
+        assert_eq!(expected, actual);
+    }
+
+    #[test]
+    fn parse_logs() {
+        let expected = {
+            let mut b = PolicyBuilder::new();
+            b.add_log(
+                hex!("fd07e34679d68f7042f0d2d3d21e956abdd0b56b1abc3d659b2b51f8e40a113e").into(),
+                None,
+            )
+            .unwrap();
+            b.add_log(
+                hex!("348129817da7b3cbf1f87e0d82b65bc300869e4b65ba40f0f0c23588c5279d2b").into(),
+                Some("http://example.com".into()),
+            )
+            .unwrap();
+            b.build()
+        };
+        let actual =
+            parse_policy("\
+                log fd07e34679d68f7042f0d2d3d21e956abdd0b56b1abc3d659b2b51f8e40a113e\n\
+                log 348129817da7b3cbf1f87e0d82b65bc300869e4b65ba40f0f0c23588c5279d2b http://example.com\n\
+                quorum none
+                ").unwrap();
+        assert_eq!(expected, actual);
+    }
+
+    #[test]
+    fn parse_witnesses() {
+        let expected = {
+            let mut b = PolicyBuilder::new();
+            b.add_witness(
+                "WIT-1".into(),
+                hex!("acf9f514e85ba44ebf00faf1bc36a16c42f2915c7b2728b489223fdcf3f2b6e3").into(),
+                None,
+            )
+            .unwrap();
+            b.add_witness(
+                "WIT-2".into(),
+                hex!("df329030f76b3616f1c50f3f8ae7ce6cf3fa92905ab7ce47bbc8be71226b65c9").into(),
+                Some("http://example.com/witness".into()),
+            )
+            .unwrap();
+            b.build()
+        };
+        let actual = parse_policy("\
+                witness WIT-1 acf9f514e85ba44ebf00faf1bc36a16c42f2915c7b2728b489223fdcf3f2b6e3\n\
+                witness WIT-2 df329030f76b3616f1c50f3f8ae7ce6cf3fa92905ab7ce47bbc8be71226b65c9 http://example.com/witness\n\
+                quorum none\n\
+                ").unwrap();
+        assert_eq!(expected, actual);
+    }
+
+    #[test]
+    fn parse_group() {
+        let expected = {
+            let mut b = PolicyBuilder::new();
+            b.add_witness(
+                "WIT-1".into(),
+                hex!("acf9f514e85ba44ebf00faf1bc36a16c42f2915c7b2728b489223fdcf3f2b6e3").into(),
+                None,
+            )
+            .unwrap();
+            b.add_witness(
+                "WIT-2".into(),
+                hex!("df329030f76b3616f1c50f3f8ae7ce6cf3fa92905ab7ce47bbc8be71226b65c9").into(),
+                Some("http://example.com/witness".into()),
+            )
+            .unwrap();
+            b.add_group("GRP".into(), 1, vec!["WIT-1".into(), "WIT-2".into()])
+                .unwrap();
+            b.set_quorum("GRP".into()).unwrap();
+            b.build()
+        };
+        let actual = parse_policy("\
+                witness WIT-1 acf9f514e85ba44ebf00faf1bc36a16c42f2915c7b2728b489223fdcf3f2b6e3\n\
+                witness WIT-2 df329030f76b3616f1c50f3f8ae7ce6cf3fa92905ab7ce47bbc8be71226b65c9 http://example.com/witness\n\
+                group GRP 1 WIT-1 WIT-2\n\
+                quorum GRP\n\
+                ").unwrap();
+        assert_eq!(expected, actual);
+    }
+
+    #[test]
+    fn parse_policy_invalid_public_key() {
+        insta::assert_snapshot!(
+            parse_policy("log fd07e34679d68f7042f0d2d3d21e956abdd0b56b1abc3d659b2b51f8e40a11").unwrap_err(),
+            @"line 1: invalid public key: wrong length",
+        );
+        insta::assert_snapshot!(
+            parse_policy("log xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx").unwrap_err(),
+            @"line 1: invalid public key: invalid hex encoding",
+        );
+    }
+
+    #[test]
+    fn parse_policy_invalid_log_rule() {
+        insta::assert_snapshot!(
+            parse_policy("log").unwrap_err(),
+            @"line 1: invalid log rule: expected 1 or 2 arguments, got 0",
+        );
+        insta::assert_snapshot!(
+            parse_policy("log fd07e34679d68f7042f0d2d3d21e956abdd0b56b1abc3d659b2b51f8e40a113e https://log.io/ xxx").unwrap_err(),
+            @"line 1: invalid log rule: expected 1 or 2 arguments, got 3",
+        );
+    }
+
+    #[test]
+    fn parse_policy_invalid_witness_rule() {
+        insta::assert_snapshot!(
+            parse_policy("witness foo").unwrap_err(),
+            @"line 1: invalid witness rule: expected 2 or 3 arguments, got 1",
+        );
+        insta::assert_snapshot!(
+            parse_policy("witness foo fd07e34679d68f7042f0d2d3d21e956abdd0b56b1abc3d659b2b51f8e40a113e https://log.io/ xxx").unwrap_err(),
+            @"line 1: invalid witness rule: expected 2 or 3 arguments, got 4",
+        );
+    }
+
+    #[test]
+    fn parse_policy_invalid_group_rule() {
+        insta::assert_snapshot!(
+            parse_policy("group foo any").unwrap_err(),
+            @"line 1: invalid group rule: expected at least 3 arguments, got 2",
+        );
+    }
+
+    #[test]
+    fn parse_policy_invalid_quorum_rule() {
+        insta::assert_snapshot!(
+            parse_policy("quorum").unwrap_err(),
+            @"line 1: invalid quorum rule: expected exactly one argument",
+        );
+        insta::assert_snapshot!(
+            parse_policy("quorum foo bar").unwrap_err(),
+            @"line 1: invalid quorum rule: expected exactly one argument",
+        );
+        insta::assert_snapshot!(
+            parse_policy("quorum foo").unwrap_err(),
+            @"line 1: invalid quorum rule: foo: no sutch witness",
+        );
+    }
+
+    #[test]
+    fn parse_policy_multiple_quorum() {
+        insta::assert_snapshot!(
+            parse_policy("\
+                witness WIT-1 acf9f514e85ba44ebf00faf1bc36a16c42f2915c7b2728b489223fdcf3f2b6e3\n\
+                witness WIT-2 df329030f76b3616f1c50f3f8ae7ce6cf3fa92905ab7ce47bbc8be71226b65c9\n\
+                quorum WIT-1\n\
+                quorum WIT-2\n\
+            ").unwrap_err(),
+            @"line 4: invalid quorum rule: quorum already set",
+        );
+    }
+
+    #[test]
+    fn parse_policy_unknown_keyword() {
+        insta::assert_snapshot!(parse_policy("foo bar").unwrap_err(), @"line 1: unknown keyword `foo`")
+    }
+}