diff --git a/src/deflate.rs b/src/deflate.rs
index 3b5b735..4cd0a3e 100644
--- a/src/deflate.rs
+++ b/src/deflate.rs
@@ -1,5 +1,5 @@
 use alloc::vec::Vec;
-use core::{cmp, iter};
+use core::{cmp, iter, num::NonZeroU64};
 
 use log::{debug, log_enabled};
 
@@ -1186,7 +1186,14 @@ fn blocksplit_attempt<W: Write>(
     for &item in &splitpoints_uncompressed {
         let mut s = ZopfliBlockState::new(options, last, item);
 
-        let store = lz77_optimal(&mut s, in_data, last, item, options.iteration_count.get());
+        let store = lz77_optimal(
+            &mut s,
+            in_data,
+            last,
+            item,
+            options.iteration_count.map(NonZeroU64::get),
+            options.iterations_without_improvement.map(NonZeroU64::get),
+        );
         totalcost += calculate_block_size_auto_type(&store, 0, store.size());
 
         // ZopfliAppendLZ77Store(&store, &lz77);
@@ -1202,7 +1209,14 @@ fn blocksplit_attempt<W: Write>(
 
     let mut s = ZopfliBlockState::new(options, last, inend);
 
-    let store = lz77_optimal(&mut s, in_data, last, inend, options.iteration_count.get());
+    let store = lz77_optimal(
+        &mut s,
+        in_data,
+        last,
+        inend,
+        options.iteration_count.map(NonZeroU64::get),
+        options.iterations_without_improvement.map(NonZeroU64::get),
+    );
     totalcost += calculate_block_size_auto_type(&store, 0, store.size());
 
     // ZopfliAppendLZ77Store(&store, &lz77);
diff --git a/src/lib.rs b/src/lib.rs
index ddac158..baa7c1a 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -60,7 +60,7 @@ mod util;
 #[cfg(feature = "zlib")]
 mod zlib;
 
-use core::num::NonZeroU8;
+use core::num::NonZeroU64;
 #[cfg(all(not(doc), feature = "std"))]
 use std::io::{Error, Write};
 
@@ -80,10 +80,13 @@ pub struct Options {
     #[cfg_attr(
         all(test, feature = "std"),
         proptest(
-            strategy = "(1..=10u8).prop_map(|iteration_count| NonZeroU8::new(iteration_count).unwrap())"
+            strategy = "(1..=10u64).prop_map(|iteration_count| NonZeroU64::new(iteration_count))"
         )
     )]
-    pub iteration_count: NonZeroU8,
+    pub iteration_count: Option<NonZeroU64>,
+    /// Stop after rerunning forward and backward pass this many times without finding
+    /// a smaller representation of the block.
+    pub iterations_without_improvement: Option<NonZeroU64>,
     /// Maximum amount of blocks to split into (0 for unlimited, but this can give
     /// extreme results that hurt compression on some files).
     ///
@@ -94,7 +97,8 @@ pub struct Options {
 impl Default for Options {
     fn default() -> Options {
         Options {
-            iteration_count: NonZeroU8::new(15).unwrap(),
+            iteration_count: Some(NonZeroU64::new(15).unwrap()),
+            iterations_without_improvement: None,
             maximum_block_splits: 15,
         }
     }
diff --git a/src/squeeze.rs b/src/squeeze.rs
index 04b90f7..f04de97 100644
--- a/src/squeeze.rs
+++ b/src/squeeze.rs
@@ -450,7 +450,8 @@ pub fn lz77_optimal<C: Cache>(
     in_data: &[u8],
     instart: usize,
     inend: usize,
-    numiterations: u8,
+    max_iterations: Option<u64>,
+    max_iterations_without_improvement: Option<u64>,
 ) -> Lz77Store {
     /* Dist to get to here with smallest cost. */
     let mut currentstore = Lz77Store::new();
@@ -470,13 +471,15 @@ pub fn lz77_optimal<C: Cache>(
     let mut lastcost = 0.0;
     /* Try randomizing the costs a bit once the size stabilizes. */
     let mut ran_state = RanState::new();
-    let mut lastrandomstep = -1;
+    let mut lastrandomstep = u64::MAX;
 
     /* Do regular deflate, then loop multiple shortest path runs, each time using
     the statistics of the previous run. */
     /* Repeat statistics with each time the cost model from the previous stat
     run. */
-    for i in 0..numiterations as i32 {
+    let mut current_iteration: u64 = 0;
+    let mut iterations_without_improvement: u64 = 0;
+    loop {
         currentstore.reset();
         lz77_optimal_run(
             s,
@@ -491,30 +494,43 @@ pub fn lz77_optimal<C: Cache>(
         let cost = calculate_block_size(&currentstore, 0, currentstore.size(), BlockType::Dynamic);
 
         if cost < bestcost {
+            iterations_without_improvement = 0;
             /* Copy to the output store. */
             outputstore = currentstore.clone();
             beststats = stats;
             bestcost = cost;
 
-            debug!("Iteration {}: {} bit", i, cost);
+            debug!("Iteration {}: {} bit", current_iteration, cost);
         } else {
-            trace!("Iteration {}: {} bit", i, cost);
+            iterations_without_improvement += 1;
+            trace!("Iteration {}: {} bit", current_iteration, cost);
+            if let Some(max_iterations_without_improvement) = max_iterations_without_improvement {
+                if iterations_without_improvement >= max_iterations_without_improvement {
+                    break;
+                }
+            }
+        }
+        current_iteration += 1;
+        if let Some(max_iterations) = max_iterations {
+            if current_iteration >= max_iterations {
+                break;
+            }
         }
         let laststats = stats;
         stats.clear_freqs();
         stats.get_statistics(&currentstore);
-        if lastrandomstep != -1 {
+        if lastrandomstep != u64::MAX {
             /* This makes it converge slower but better. Do it only once the
             randomness kicks in so that if the user does few iterations, it gives a
             better result sooner. */
             stats = add_weighed_stat_freqs(&stats, 1.0, &laststats, 0.5);
             stats.calculate_entropy();
         }
-        if i > 5 && (cost - lastcost).abs() < f64::EPSILON {
+        if current_iteration > 5 && (cost - lastcost).abs() < f64::EPSILON {
             stats = beststats;
             stats.randomize_stat_freqs(&mut ran_state);
             stats.calculate_entropy();
-            lastrandomstep = i;
+            lastrandomstep = current_iteration;
         }
         lastcost = cost;
     }