diff --git a/Outputs/OpenGL/ScanTarget.cpp b/Outputs/OpenGL/ScanTarget.cpp
index 83fe1613b..846bed6ac 100644
--- a/Outputs/OpenGL/ScanTarget.cpp
+++ b/Outputs/OpenGL/ScanTarget.cpp
@@ -131,6 +131,8 @@ void ScanTarget::set_modals(Modals modals) {
 Outputs::Display::ScanTarget::Scan *ScanTarget::begin_scan() {
 	if(allocation_has_failed_) return nullptr;
 
+	std::lock_guard<std::mutex> lock_guard(write_pointers_mutex_);
+
 	const auto result = &scan_buffer_[write_pointers_.scan_buffer];
 	const auto read_pointers = read_pointers_.load();
 
@@ -154,6 +156,7 @@ Outputs::Display::ScanTarget::Scan *ScanTarget::begin_scan() {
 
 void ScanTarget::end_scan() {
 	if(vended_scan_) {
+		std::lock_guard<std::mutex> lock_guard(write_pointers_mutex_);
 		vended_scan_->data_y = TextureAddressGetY(vended_write_area_pointer_);
 		vended_scan_->line = write_pointers_.line;
 		vended_scan_->scan.end_points[0].data_offset += TextureAddressGetX(vended_write_area_pointer_);
@@ -182,6 +185,8 @@ uint8_t *ScanTarget::begin_data(size_t required_length, size_t required_alignmen
 		return nullptr;
 	}
 
+	std::lock_guard<std::mutex> lock_guard(write_pointers_mutex_);
+
 	// Determine where the proposed write area would start and end.
 	uint16_t output_y = TextureAddressGetY(write_pointers_.write_area);
 
@@ -222,6 +227,8 @@ uint8_t *ScanTarget::begin_data(size_t required_length, size_t required_alignmen
 void ScanTarget::end_data(size_t actual_length) {
 	if(allocation_has_failed_ || !data_is_allocated_) return;
 
+	std::lock_guard<std::mutex> lock_guard(write_pointers_mutex_);
+
 	// Bookend the start of the new data, to safeguard for precision errors in sampling.
 	memcpy(
 		&write_area_texture_[size_t(write_pointers_.write_area - 1) * data_type_size_],
@@ -268,6 +275,7 @@ void ScanTarget::announce(Event event, bool is_visible, const Outputs::Display::
 	if(output_is_visible_ == is_visible) return;
 	if(is_visible) {
 		const auto read_pointers = read_pointers_.load();
+		std::lock_guard<std::mutex> lock_guard(write_pointers_mutex_);
 
 		// Commit the most recent line only if any scans fell on it.
 		// Otherwise there's no point outputting it, it'll contribute nothing.
@@ -336,14 +344,20 @@ void ScanTarget::announce(Event event, bool is_visible, const Outputs::Display::
 
 void ScanTarget::setup_pipeline() {
 	const auto data_type_size = Outputs::Display::size_for_data_type(modals_.input_data_type);
-	if(data_type_size != data_type_size_) {
-		// TODO: flush output.
 
-		data_type_size_ = data_type_size;
-		write_area_texture_.resize(WriteAreaWidth*WriteAreaHeight*data_type_size_);
+	// Ensure the lock guard here has a restricted scope; this is the only time that a thread
+	// other than the main owner of write_pointers_ may adjust it.
+	{
+		std::lock_guard<std::mutex> lock_guard(write_pointers_mutex_);
+		if(data_type_size != data_type_size_) {
+			// TODO: flush output.
 
-		write_pointers_.scan_buffer = 0;
-		write_pointers_.write_area = 0;
+			data_type_size_ = data_type_size;
+			write_area_texture_.resize(WriteAreaWidth*WriteAreaHeight*data_type_size_);
+
+			write_pointers_.scan_buffer = 0;
+			write_pointers_.write_area = 0;
+		}
 	}
 
 	// Prepare to bind line shaders.
diff --git a/Outputs/OpenGL/ScanTarget.hpp b/Outputs/OpenGL/ScanTarget.hpp
index f7eeb816e..cda6c47cc 100644
--- a/Outputs/OpenGL/ScanTarget.hpp
+++ b/Outputs/OpenGL/ScanTarget.hpp
@@ -116,6 +116,10 @@ class ScanTarget: public Outputs::Display::ScanTarget {
 		/// A pointer to the next thing that should be provided to the caller for data.
 		PointerSet write_pointers_;
 
+		/// A mutex for gettng access to write_pointers_; access to write_pointers_
+		/// or data_type_size_ is almost never contended, so this is cheap for the main use case.
+		std::mutex write_pointers_mutex_;
+
 		/// A pointer to the final thing currently cleared for submission.
 		std::atomic<PointerSet> submit_pointers_;