diff --git a/bulk_indexer.go b/bulk_indexer.go
index 5788f57..41c359d 100644
--- a/bulk_indexer.go
+++ b/bulk_indexer.go
@@ -78,6 +78,7 @@ type BulkIndexer struct {
 	itemsAdded         int
 	bytesFlushed       int
 	bytesUncompFlushed int
+	pendingUncomBytes  int
 	jsonw              fastjson.Writer
 	writer             *countWriter
 	gzipw              *gzip.Writer
@@ -203,22 +204,21 @@ func NewBulkIndexer(cfg BulkIndexerConfig) (*BulkIndexer, error) {
 	} else {
 		writer = &b.buf
 	}
-	b.writer = &countWriter{
-		0,
-		writer,
-	}
+	b.writer = &countWriter{0, writer}
 	return b, nil
 }
 
 // Reset resets bulk indexer, ready for a new request.
 func (b *BulkIndexer) Reset() {
 	b.bytesFlushed = 0
-	b.writer.count = 0
+	b.bytesUncompFlushed = 0
 }
 
+// resetBuf resets compressed buffer after flushing it to Elasticsearch
 func (b *BulkIndexer) resetBuf() {
 	b.itemsAdded = 0
 	b.writer.count = 0
+	b.pendingUncomBytes = 0
 	b.buf.Reset()
 	if b.gzipw != nil {
 		b.gzipw.Reset(&b.buf)
@@ -235,6 +235,11 @@ func (b *BulkIndexer) Len() int {
 	return b.buf.Len()
 }
 
+// Len returns the number of buffered bytes.
+func (b *BulkIndexer) UncompressedLen() int {
+	return b.writer.count + b.pendingUncomBytes
+}
+
 // BytesFlushed returns the number of bytes flushed by the bulk indexer.
 func (b *BulkIndexer) BytesFlushed() int {
 	return b.bytesFlushed
@@ -284,6 +289,7 @@ func (b *BulkIndexer) writeMeta(index, documentID string) {
 
 // Flush executes a bulk request if there are any items buffered, and clears out the buffer.
 func (b *BulkIndexer) Flush(ctx context.Context) (BulkIndexerResponseStat, error) {
+
 	if b.itemsAdded == 0 {
 		return BulkIndexerResponseStat{}, nil
 	}
@@ -314,7 +320,7 @@ func (b *BulkIndexer) Flush(ctx context.Context) (BulkIndexerResponseStat, error
 	}
 
 	bytesFlushed := b.buf.Len()
-	bytesUncompFlushed := b.writer.count
+	bytesUncompFlushed := b.writer.count + b.pendingUncomBytes
 	res, err := req.Do(ctx, b.config.Client)
 	if err != nil {
 		b.resetBuf()
@@ -409,7 +415,7 @@ func (b *BulkIndexer) Flush(ctx context.Context) (BulkIndexerResponseStat, error
 					if len(buf) == 0 {
 						n, err := gr.Read(buf[:cap(buf)])
 						if err != nil && err != io.EOF {
-							return resp, fmt.Errorf("failed to read from compressed buffer: %w", err)
+							return resp, fmt.Errorf("failed to read from uncompressed buffer: %w", err)
 						}
 						buf = buf[:n]
 					}
@@ -422,7 +428,7 @@ func (b *BulkIndexer) Flush(ctx context.Context) (BulkIndexerResponseStat, error
 						seen += newlines
 						n, err := gr.Read(buf[:cap(buf)])
 						if err != nil && err != io.EOF {
-							return resp, fmt.Errorf("failed to read from compressed buffer: %w", err)
+							return resp, fmt.Errorf("failed to read from uncompressed buffer: %w", err)
 						}
 						buf = buf[:n]
 						newlines = bytes.Count(buf, []byte{'\n'})
@@ -441,7 +447,7 @@ func (b *BulkIndexer) Flush(ctx context.Context) (BulkIndexerResponseStat, error
 							seen += newlines
 							n, err := gr.Read(buf[:cap(buf)])
 							if err != nil && err != io.EOF {
-								return resp, fmt.Errorf("failed to read from compressed buffer: %w", err)
+								return resp, fmt.Errorf("failed to read from uncompressed buffer: %w", err)
 							}
 							buf = buf[:n]
 							newlines = bytes.Count(buf, []byte{'\n'})
@@ -480,6 +486,11 @@ func (b *BulkIndexer) Flush(ctx context.Context) (BulkIndexerResponseStat, error
 		if len(tmp) > 0 {
 			resp.FailedDocs = tmp
 		}
+
+		// record uncompressed bytes written to b.buf
+		// this should be added to the bytesUncompFlushed on next flush
+		b.pendingUncomBytes = b.writer.count
+		b.writer.count = 0
 	}
 
 	return resp, nil
diff --git a/bulk_indexer_test.go b/bulk_indexer_test.go
index 3ee05a8..d6e0c18 100644
--- a/bulk_indexer_test.go
+++ b/bulk_indexer_test.go
@@ -79,11 +79,16 @@ func TestBulkIndexer(t *testing.T) {
 
 			itemCount := 1_000
 			generateLoad(itemCount)
-
-			// All items should be successfully flushed
+			uncompressed := indexer.UncompressedLen()
+			uncompressedDocSize := uncompressed / itemCount
 			stat, err := indexer.Flush(context.Background())
 			require.NoError(t, err)
 			require.Equal(t, int64(itemCount), stat.Indexed)
+			require.Equal(t, uncompressed, indexer.BytesUncompFlushed())
+
+			// nothing is in the buffer if all succeeded
+			require.Equal(t, 0, indexer.Len())
+			require.Equal(t, 0, indexer.UncompressedLen())
 
 			// Simulate ES failure, all items should be enqueued for retries
 			esFailing.Store(true)
@@ -97,17 +102,26 @@ func TestBulkIndexer(t *testing.T) {
 				require.Equal(t, itemCount, len(stat.FailedDocs))
 				require.Equal(t, int64(itemCount), stat.RetriedDocs)
 
+				// all the flushed bytes are now in the buffer again to be retried
+				require.Equal(t, indexer.UncompressedLen(), indexer.BytesUncompFlushed())
 				// Generate more load, all these items should be enqueued for retries
 				generateLoad(10)
 				itemCount += 10
 				require.Equal(t, itemCount, indexer.Items())
+				expectedBufferedSize := indexer.BytesUncompFlushed() + (10 * uncompressedDocSize)
+				require.Equal(t, expectedBufferedSize, indexer.UncompressedLen())
 			}
 
+			uncompressedSize := indexer.UncompressedLen()
 			// Recover ES and ensure all items are indexed
 			esFailing.Store(false)
 			stat, err = indexer.Flush(context.Background())
 			require.NoError(t, err)
 			require.Equal(t, int64(itemCount), stat.Indexed)
+			require.Equal(t, uncompressedSize, indexer.BytesUncompFlushed())
+			// no documents to retry so buffer is empty
+			require.Equal(t, 0, indexer.Len())
+			require.Equal(t, 0, indexer.UncompressedLen())
 		})
 	}
 }