diff --git a/crates/typst/src/foundations/float.rs b/crates/typst/src/foundations/float.rs
index fd094950b..ff3e62a50 100644
--- a/crates/typst/src/foundations/float.rs
+++ b/crates/typst/src/foundations/float.rs
@@ -2,7 +2,10 @@ use std::num::ParseFloatError;
 
 use ecow::{eco_format, EcoString};
 
-use crate::foundations::{cast, func, repr, scope, ty, Repr, Str};
+use crate::diag::StrResult;
+use crate::foundations::{
+    bail, cast, func, repr, scope, ty, Bytes, Endianness, Repr, Str,
+};
 use crate::layout::Ratio;
 
 /// A floating-point number.
@@ -106,6 +109,58 @@ impl f64 {
     pub fn signum(self) -> f64 {
         f64::signum(self)
     }
+
+    /// Converts bytes to a float.
+    ///
+    /// ```example
+    /// #float.from-bytes(bytes((0, 0, 0, 0, 0, 0, 240, 63))) \
+    /// #float.from-bytes(bytes((63, 240, 0, 0, 0, 0, 0, 0)), endian: "big")
+    /// ```
+    #[func]
+    pub fn from_bytes(
+        /// The bytes that should be converted to a float.
+        ///
+        /// Must be of length exactly 8 so that the result fits into a 64-bit
+        /// float.
+        bytes: Bytes,
+        /// The endianness of the conversion.
+        #[named]
+        #[default(Endianness::Little)]
+        endian: Endianness,
+    ) -> StrResult<f64> {
+        // Convert slice to an array of length 8.
+        let buf: [u8; 8] = match bytes.as_ref().try_into() {
+            Ok(buffer) => buffer,
+            Err(_) => bail!("bytes must have a length of exactly 8"),
+        };
+
+        Ok(match endian {
+            Endianness::Little => f64::from_le_bytes(buf),
+            Endianness::Big => f64::from_be_bytes(buf),
+        })
+    }
+
+    /// Converts a float to bytes.
+    ///
+    /// ```example
+    /// #array(1.0.to-bytes(endian: "big")) \
+    /// #array(1.0.to-bytes())
+    /// ```
+    #[func]
+    pub fn to_bytes(
+        self,
+        /// The endianness of the conversion.
+        #[named]
+        #[default(Endianness::Little)]
+        endian: Endianness,
+    ) -> Bytes {
+        match endian {
+            Endianness::Little => self.to_le_bytes(),
+            Endianness::Big => self.to_be_bytes(),
+        }
+        .as_slice()
+        .into()
+    }
 }
 
 impl Repr for f64 {
diff --git a/crates/typst/src/foundations/int.rs b/crates/typst/src/foundations/int.rs
index 4c1335fc9..79a8a86f9 100644
--- a/crates/typst/src/foundations/int.rs
+++ b/crates/typst/src/foundations/int.rs
@@ -341,12 +341,15 @@ impl Repr for i64 {
     }
 }
 
-/// Represents the byte order used for converting integers to bytes and vice versa.
+/// Represents the byte order used for converting integers and floats to bytes
+/// and vice versa.
 #[derive(Debug, Copy, Clone, Eq, PartialEq, Hash, Cast)]
 pub enum Endianness {
-    /// Big-endian byte order: the highest-value byte is at the beginning of the bytes.
+    /// Big-endian byte order: The highest-value byte is at the beginning of the
+    /// bytes.
     Big,
-    /// Little-endian byte order: the lowest-value byte is at the beginning of the bytes.
+    /// Little-endian byte order: The lowest-value byte is at the beginning of
+    /// the bytes.
     Little,
 }
 
diff --git a/tests/suite/foundations/float.typ b/tests/suite/foundations/float.typ
index 01769260e..67d4acbf7 100644
--- a/tests/suite/foundations/float.typ
+++ b/tests/suite/foundations/float.typ
@@ -42,6 +42,17 @@
 #test(float(-calc.inf).signum(), -1.0)
 #test(float(float.nan).signum().is-nan(), true)
 
+--- float-from-and-to-bytes ---
+// Test float `from-bytes()` and `to-bytes()`.
+#test(float.from-bytes(bytes((0, 0, 0, 0, 0, 0, 240, 63))), 1.0)
+#test(float.from-bytes(bytes((63, 240, 0, 0, 0, 0, 0, 0)), endian: "big"), 1.0)
+#test(1.0.to-bytes(), bytes((0, 0, 0, 0, 0, 0, 240, 63)))
+#test(1.0.to-bytes(endian: "big"), bytes((63, 240, 0, 0, 0, 0, 0, 0)))
+
+--- float-from-bytes-bad-length ---
+// Error: 2-54 bytes must have a length of exactly 8
+#float.from-bytes(bytes((0, 0, 0, 0, 0, 0, 0, 1, 0)))
+
 --- float-repr ---
 // Test the `repr` function with floats.
 #repr(12.0) \