diff --git a/video/qt_decoder.cpp b/video/qt_decoder.cpp
index a226f9fb374..dd26b137c17 100644
--- a/video/qt_decoder.cpp
+++ b/video/qt_decoder.cpp
@@ -299,7 +299,8 @@ QuickTimeDecoder::VideoTrackHandler::VideoTrackHandler(QuickTimeDecoder *decoder
 
 	_curEdit = 0;
 	_curFrame = -1;
-	enterNewEditListEntry(true); // might set _curFrame
+	_delayedFrameToBufferTo = -1;
+	enterNewEditListEntry(true, true); // might set _curFrame
 
 	_durationOverride = -1;
 	_scaledSurface = 0;
@@ -369,6 +370,8 @@ bool QuickTimeDecoder::VideoTrackHandler::endOfTrack() const {
 }
 
 bool QuickTimeDecoder::VideoTrackHandler::seek(const Audio::Timestamp &requestedTime) {
+	_delayedFrameToBufferTo = -1; // abort any delayed buffering
+
 	uint32 convertedFrames = requestedTime.convertToFramerate(_decoder->_timeScale).totalNumberOfFrames();
 	for (_curEdit = 0; !atLastEdit(); _curEdit++)
 		if (convertedFrames >= _parent->editList[_curEdit].timeOffset && convertedFrames < _parent->editList[_curEdit].timeOffset + _parent->editList[_curEdit].trackDuration)
@@ -564,6 +567,8 @@ const byte *QuickTimeDecoder::VideoTrackHandler::getPalette() const {
 }
 
 bool QuickTimeDecoder::VideoTrackHandler::setReverse(bool reverse) {
+	_delayedFrameToBufferTo = -1; // abort any delayed buffering
+
 	_reversed = reverse;
 
 	if (_reversed) {
@@ -705,7 +710,7 @@ bool QuickTimeDecoder::VideoTrackHandler::isEmptyEdit() const {
 	return (_parent->editList[_curEdit].mediaTime == -1);
 }
 
-void QuickTimeDecoder::VideoTrackHandler::enterNewEditListEntry(bool bufferFrames) {
+void QuickTimeDecoder::VideoTrackHandler::enterNewEditListEntry(bool bufferFrames, bool initializingTrack) {
 	if (atLastEdit())
 		return;
 
@@ -749,8 +754,15 @@ void QuickTimeDecoder::VideoTrackHandler::enterNewEditListEntry(bool bufferFrame
 		// Track down the keyframe
 		// Then decode until the frame before target
 		_curFrame = findKeyFrame(frameNum) - 1;
-		while (_curFrame < (int32)frameNum - 1)
-			bufferNextFrame();
+		if (initializingTrack) {
+			// We can't decode frames during track initialization,
+			// so delay buffering until the first decode.
+			_delayedFrameToBufferTo = (int32)frameNum - 1;
+		} else {
+			while (_curFrame < (int32)frameNum - 1) {
+				bufferNextFrame();
+			}
+		}
 	} else {
 		// Since frameNum is the frame that needs to be displayed
 		// we'll set _curFrame to be the "last frame displayed"
@@ -761,6 +773,16 @@ void QuickTimeDecoder::VideoTrackHandler::enterNewEditListEntry(bool bufferFrame
 }
 
 const Graphics::Surface *QuickTimeDecoder::VideoTrackHandler::bufferNextFrame() {
+	// Buffer any frames that were identified during track initialization
+	// and delayed until decoding.
+	if (_delayedFrameToBufferTo != -1) {
+		int32 frameNum = _delayedFrameToBufferTo;
+		_delayedFrameToBufferTo = -1;
+		while (_curFrame < frameNum) {
+			bufferNextFrame();
+		}
+	}
+
 	_curFrame++;
 
 	// Get the next packet
diff --git a/video/qt_decoder.h b/video/qt_decoder.h
index 62496629e1f..581949cca09 100644
--- a/video/qt_decoder.h
+++ b/video/qt_decoder.h
@@ -154,6 +154,7 @@ private:
 		Common::QuickTimeParser::Track *_parent;
 		uint32 _curEdit;
 		int32 _curFrame;
+		int32 _delayedFrameToBufferTo;
 		uint32 _nextFrameStartTime; // media time
 		Graphics::Surface *_scaledSurface;
 		int32 _durationOverride;    // media time
@@ -171,7 +172,7 @@ private:
 		uint32 getCurFrameDuration();            // media time
 		uint32 findKeyFrame(uint32 frame) const;
 		bool isEmptyEdit() const;
-		void enterNewEditListEntry(bool bufferFrames);
+		void enterNewEditListEntry(bool bufferFrames, bool intializingTrack = false);
 		const Graphics::Surface *bufferNextFrame();
 		uint32 getRateAdjustedFrameTime() const; // media time
 		uint32 getCurEditTimeOffset() const;     // media time