diff --git a/src/page/manager.rs b/src/page/manager.rs
index 21996001..2fdcc65d 100644
--- a/src/page/manager.rs
+++ b/src/page/manager.rs
@@ -1,5 +1,6 @@
 use crate::page::PageId;
 
+mod dirty;
 pub(super) mod mmap;
 pub(super) mod options;
 
diff --git a/src/page/manager/dirty.rs b/src/page/manager/dirty.rs
new file mode 100644
index 00000000..f50417e9
--- /dev/null
+++ b/src/page/manager/dirty.rs
@@ -0,0 +1,243 @@
+use std::collections::{BTreeMap, HashSet};
+
+use crate::page::PageId;
+
+/// A container which tracks dirty pages in a memory mapped file.
+#[derive(Debug, Default)]
+pub(crate) struct DirtyPages {
+    // A set of reallocated dirty pages which need to be flushed to disk.
+    set: HashSet<PageId>,
+    // Half-open ranges [start, end) of dirty pages, non-overlapping and non-adjacent.
+    runs: BTreeMap<PageId, PageId>,
+}
+
+impl DirtyPages {
+    /// Iterate over the dirty page runs as (byte offset, length) pairs.
+    pub(crate) fn byte_runs<'a>(&'a self) -> impl Iterator<Item = (usize, usize)> + 'a {
+        self.runs
+            .iter()
+            .map(|(&start, &end)| (start.as_offset(), end.as_offset() - start.as_offset()))
+    }
+
+    /// Marks a page as dirty, returning true if the page was newly marked.
+    /// This attempts to find
+    pub(crate) fn mark_dirty(&mut self, page_id: PageId) -> bool {
+        if !self.set.insert(page_id) {
+            return false;
+        }
+
+        let next_page_id = page_id.inc();
+
+        // Find predecessor run: end <= id
+        let predecessor = self.runs.range(..=page_id).next_back();
+        let join_left = predecessor.is_some_and(|(_, &end)| end == page_id);
+
+        // Find successor run: start > id
+        let successor =
+            next_page_id.and_then(|next_page_id| self.runs.range(next_page_id..).next());
+        let join_right = successor.is_some_and(|(&start, _)| start == next_page_id.unwrap());
+
+        match (join_left, join_right) {
+            (true, true) => {
+                // Join predecessor and successor runs into a single run
+                let (&l_start, _) = predecessor.unwrap();
+                let (&r_start, &r_end) = successor.unwrap();
+                self.runs.remove(&l_start);
+                self.runs.remove(&r_start);
+                self.runs.insert(l_start, r_end);
+            }
+            (true, false) => {
+                // Extend predecessor run to include id
+                let (&l_start, _) = predecessor.unwrap();
+                let l_end = self.runs.get_mut(&l_start).unwrap();
+                *l_end = next_page_id.unwrap();
+            }
+            (false, true) => {
+                // Extend successor run backwards to include id
+                let (&r_start, &r_end) = successor.unwrap();
+                self.runs.remove(&r_start);
+                self.runs.insert(page_id, r_end);
+            }
+            (false, false) => {
+                // Create new run containing only the id
+                self.runs.insert(page_id, next_page_id.unwrap());
+            }
+        }
+
+        true
+    }
+
+    /// Clears the dirty pages set and runs.
+    pub(crate) fn clear(&mut self) {
+        self.set.clear();
+        self.runs.clear();
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use proptest::prelude::*;
+
+    use crate::page_id;
+
+    use super::*;
+
+    #[test]
+    fn test_clear() {
+        let mut dirty_pages = DirtyPages::default();
+        assert!(dirty_pages.mark_dirty(page_id!(1)));
+        assert!(!dirty_pages.mark_dirty(page_id!(1)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 4096)]);
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(1)));
+        assert!(!dirty_pages.mark_dirty(page_id!(1)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 4096)]);
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+    }
+
+    #[test]
+    fn test_mark_dirty_noncontiguous_runs() {
+        let mut dirty_pages = DirtyPages::default();
+
+        assert!(dirty_pages.mark_dirty(page_id!(1)));
+        assert!(!dirty_pages.mark_dirty(page_id!(1)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(3)));
+        assert!(!dirty_pages.mark_dirty(page_id!(3)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 4096), (8192, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(11)));
+        assert_eq!(
+            dirty_pages.byte_runs().collect::<Vec<_>>(),
+            vec![(0, 4096), (8192, 4096), (40960, 4096)]
+        );
+
+        assert!(dirty_pages.mark_dirty(page_id!(101)));
+        assert_eq!(
+            dirty_pages.byte_runs().collect::<Vec<_>>(),
+            vec![(0, 4096), (8192, 4096), (40960, 4096), (409600, 4096)]
+        );
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+    }
+
+    #[test]
+    fn test_mark_dirty_extend_right() {
+        let mut dirty_pages = DirtyPages::default();
+
+        assert!(dirty_pages.mark_dirty(page_id!(1)));
+        assert!(!dirty_pages.mark_dirty(page_id!(1)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(2)));
+        assert!(!dirty_pages.mark_dirty(page_id!(2)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 8192)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(3)));
+        assert!(!dirty_pages.mark_dirty(page_id!(3)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 12288)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(11)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 12288), (40960, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(12)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 12288), (40960, 8192)]);
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+    }
+
+    #[test]
+    fn test_mark_dirty_extend_left() {
+        let mut dirty_pages = DirtyPages::default();
+
+        assert!(dirty_pages.mark_dirty(page_id!(100)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(405504, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(99)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(401408, 8192)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(98)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(397312, 12288)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(2)));
+        assert_eq!(
+            dirty_pages.byte_runs().collect::<Vec<_>>(),
+            vec![(4096, 4096), (397312, 12288)]
+        );
+
+        assert!(dirty_pages.mark_dirty(page_id!(1)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(0, 8192), (397312, 12288)]);
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+    }
+
+    #[test]
+    fn test_mark_dirty_extend_both() {
+        let mut dirty_pages = DirtyPages::default();
+
+        assert!(dirty_pages.mark_dirty(page_id!(100)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(405504, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(99)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(401408, 8192)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(101)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(401408, 12288)]);
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+    }
+
+    #[test]
+    fn test_mark_dirty_merge_runs() {
+        let mut dirty_pages = DirtyPages::default();
+
+        assert!(dirty_pages.mark_dirty(page_id!(100)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(405504, 4096)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(98)));
+        assert_eq!(
+            dirty_pages.byte_runs().collect::<Vec<_>>(),
+            vec![(397312, 4096), (405504, 4096)]
+        );
+
+        assert!(dirty_pages.mark_dirty(page_id!(99)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(397312, 12288)]);
+
+        assert!(dirty_pages.mark_dirty(page_id!(102)));
+        assert_eq!(
+            dirty_pages.byte_runs().collect::<Vec<_>>(),
+            vec![(397312, 12288), (413696, 4096)]
+        );
+
+        assert!(dirty_pages.mark_dirty(page_id!(101)));
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![(397312, 20480)]);
+
+        dirty_pages.clear();
+        assert_eq!(dirty_pages.byte_runs().collect::<Vec<_>>(), vec![]);
+    }
+
+    proptest! {
+        #[test]
+        fn test_mark_dirty_random(page_ids in prop::collection::vec(any::<PageId>(), 1..1000)) {
+            let mut dirty_pages = DirtyPages::default();
+            for page_id in page_ids {
+                dirty_pages.mark_dirty(page_id);
+            }
+
+            let runs = dirty_pages.byte_runs().collect::<Vec<_>>();
+            assert!(runs.is_sorted_by_key(|(start, _)| *start));
+            // ensure that runs are non-overlapping and non-adjacent
+            assert!(runs.windows(2).all(|w| w[0].0 + w[0].1 < w[1].0));
+        }
+    }
+}
diff --git a/src/page/manager/mmap.rs b/src/page/manager/mmap.rs
index 41d62087..b03311d2 100644
--- a/src/page/manager/mmap.rs
+++ b/src/page/manager/mmap.rs
@@ -1,5 +1,6 @@
 use crate::{
     page::{
+        manager::dirty::DirtyPages,
         state::{PageState, RawPageState},
         Page, PageError, PageId, PageManagerOptions, PageMut,
     },
@@ -21,6 +22,7 @@ pub struct PageManager {
     file: Mutex<File>,
     file_len: AtomicU64,
     page_count: AtomicU32,
+    dirty_pages: Mutex<DirtyPages>,
 }
 
 impl PageManager {
@@ -104,6 +106,7 @@ impl PageManager {
             file: Mutex::new(file),
             file_len: AtomicU64::new(file_len),
             page_count: AtomicU32::new(opts.page_count),
+            dirty_pages: Mutex::new(DirtyPages::default()),
         })
     }
 
@@ -216,6 +219,8 @@ impl PageManager {
         // SAFETY: We have checked that the page fits inside the memory map.
         let data = unsafe { self.mmap.as_mut_ptr().byte_add(offset).cast() };
 
+        self.mark_dirty(page_id);
+
         // TODO: This is actually unsafe, as it's possible to call `get()` arbitrary times before
         // calling this function (this will be fixed in a future commit).
         unsafe { PageMut::from_ptr(page_id, snapshot_id, data) }
@@ -237,6 +242,8 @@ impl PageManager {
         // SAFETY: We have checked that the page fits inside the memory map.
         let data = unsafe { self.mmap.as_mut_ptr().byte_add(offset).cast() };
 
+        self.mark_dirty(page_id);
+
         // SAFETY:
         // - This is a newly created page at the end of the file, so we're guaranteed to have
         //   exclusive access to it. Even if another thread was calling `allocate()` at the same
@@ -265,10 +272,22 @@ impl PageManager {
         Ok(matches!(state.load(), PageState::Dirty(_)))
     }
 
-    /// Syncs pages to the backing file.
+    /// Adds a page to the dirty set, to be flushed to disk on next sync.
+    #[inline]
+    fn mark_dirty(&self, page_id: PageId) {
+        self.dirty_pages.lock().mark_dirty(page_id);
+    }
+
+    /// Syncs dirty pages to the backing file.
+    /// This performs up to N+1 flushes, where N is the number of pages in the dirty set.
     pub fn sync(&self) -> io::Result<()> {
         if cfg!(not(miri)) {
-            self.mmap.flush()
+            let mut dirty_pages = self.dirty_pages.lock();
+            for (offset, length) in dirty_pages.byte_runs() {
+                self.mmap.flush_range(offset, length)?;
+            }
+            dirty_pages.clear();
+            Ok(())
         } else {
             Ok(())
         }