diff --git a/noodles-cram/src/codecs/rans_4x8/encode/order_1.rs b/noodles-cram/src/codecs/rans_4x8/encode/order_1.rs
index 79915354e..4d771e414 100644
--- a/noodles-cram/src/codecs/rans_4x8/encode/order_1.rs
+++ b/noodles-cram/src/codecs/rans_4x8/encode/order_1.rs
@@ -24,34 +24,21 @@ pub fn encode(src: &[u8]) -> io::Result<Vec<u8>> {
     let mut buf = Vec::new();
     let mut states = [LOWER_BOUND; STATE_COUNT];
 
-    // The input data is split into 4 equally sized chunks.
-    let quarter = src.len() / states.len();
-
-    let chunks = [
-        &src[0..quarter],
-        &src[quarter..2 * quarter],
-        &src[2 * quarter..3 * quarter],
-        &src[3 * quarter..4 * quarter],
-    ];
-
-    // The remainder of the input buffer is processed by the last state.
-    if src.len() > 4 * quarter {
-        // The last chunk includes the last symbol of the fourth chunk (the subtraction by 1). This
-        // is safe because chunks are guaranteed to be nonempty.
-        let remainder = &src[4 * quarter - 1..];
+    let [chunk_0, chunk_1, chunk_2, chunk_3, chunk_4] = split_chunks(src);
 
-        for syms in remainder.windows(2).rev() {
-            let (i, j) = (usize::from(syms[0]), usize::from(syms[1]));
-            states[3] = state_renormalize(states[3], frequencies[i][j], &mut buf)?;
-            states[3] = state_step(states[3], frequencies[i][j], cumulative_frequencies[i][j]);
-        }
+    // § 2.2.1 "rANS entropy encoding: Interleaving" (2023-03-15): "Any remainder, when the input
+    // buffer is not divisible by 4, is processed ... by the 4th rANS state."
+    for syms in chunk_4.windows(2).rev() {
+        let (i, j) = (usize::from(syms[0]), usize::from(syms[1]));
+        states[3] = state_renormalize(states[3], frequencies[i][j], &mut buf)?;
+        states[3] = state_step(states[3], frequencies[i][j], cumulative_frequencies[i][j]);
     }
 
     let mut windows = [
-        chunks[0].windows(2).rev(),
-        chunks[1].windows(2).rev(),
-        chunks[2].windows(2).rev(),
-        chunks[3].windows(2).rev(),
+        chunk_0.windows(2).rev(),
+        chunk_1.windows(2).rev(),
+        chunk_2.windows(2).rev(),
+        chunk_3.windows(2).rev(),
     ];
 
     let mut n = 0;
@@ -68,7 +55,8 @@ pub fn encode(src: &[u8]) -> io::Result<Vec<u8>> {
         n += 1;
     }
 
-    // The last state updates are for the starting contexts, i.e., `(0, chunks[i][0])`.
+    let chunks = [chunk_0, chunk_1, chunk_2, chunk_3];
+
     for (state, chunk) in states.iter_mut().rev().zip(chunks.iter().rev()) {
         let (i, j) = (usize::from(NUL), usize::from(chunk[0]));
         *state = state_renormalize(*state, frequencies[i][j], &mut buf)?;
@@ -181,6 +169,20 @@ fn build_cumulative_frequencies(frequencies: &Frequencies) -> CumulativeFrequenc
     cumulative_frequencies
 }
 
+fn split_chunks(buf: &[u8]) -> [&[u8]; 5] {
+    let chunk_size = buf.len() / STATE_COUNT;
+
+    let (left_chunk, right_chunk) = buf.split_at(2 * chunk_size);
+    let (chunk_0, chunk_1) = left_chunk.split_at(chunk_size);
+    let (chunk_2, chunk_3_4) = right_chunk.split_at(chunk_size);
+    let (chunk_3, _) = chunk_3_4.split_at(chunk_size);
+
+    // The last chunk includes the last byte of chunk 3 for context.
+    let chunk_4 = &chunk_3_4[chunk_size - 1..];
+
+    [chunk_0, chunk_1, chunk_2, chunk_3, chunk_4]
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -282,4 +284,17 @@ mod tests {
 
         assert_eq!(normalize_frequencies(&raw_frequencies), expected);
     }
+
+    #[test]
+    fn test_split_chunks() {
+        assert_eq!(
+            split_chunks(&[1, 2, 3, 4]),
+            [&[1][..], &[2][..], &[3][..], &[4][..], &[4][..]]
+        );
+
+        assert_eq!(
+            split_chunks(&[1, 2, 3, 4, 5]),
+            [&[1][..], &[2][..], &[3][..], &[4][..], &[4, 5][..]]
+        );
+    }
 }