diff --git a/haystack/components/preprocessors/embedding_based_document_splitter.py b/haystack/components/preprocessors/embedding_based_document_splitter.py
index 17acefd3b7..34e74cfa09 100644
--- a/haystack/components/preprocessors/embedding_based_document_splitter.py
+++ b/haystack/components/preprocessors/embedding_based_document_splitter.py
@@ -445,6 +445,16 @@ def _merge_small_splits(self, splits: list[str]) -> list[str]:
         # Don't forget the last split
         merged.append(current_split)
 
+        # The loop only merges forward, so the final split can still be below min_length. Merge it backwards,
+        # subject to the same max_length limit as forward merges.
+        if (
+            len(merged) > 1
+            and len(merged[-1]) < self.min_length
+            and len(merged[-2]) + len(merged[-1]) < self.max_length
+        ):
+            trailing_split = merged.pop()
+            merged[-1] += trailing_split
+
         return merged
 
     def _split_large_splits(self, splits: list[str]) -> list[str]:
diff --git a/releasenotes/notes/merge-short-trailing-split-8f5b640fdbc3ccbe.yaml b/releasenotes/notes/merge-short-trailing-split-8f5b640fdbc3ccbe.yaml
new file mode 100644
index 0000000000..456a4d4603
--- /dev/null
+++ b/releasenotes/notes/merge-short-trailing-split-8f5b640fdbc3ccbe.yaml
@@ -0,0 +1,7 @@
+---
+fixes:
+  - |
+    Fixed ``EmbeddingBasedDocumentSplitter`` emitting a final split shorter than ``min_length``. Small splits
+    were only merged forward, so the last one had nothing left to absorb and was returned as its own document.
+    It is now merged into the preceding split, unless doing so would reach ``max_length``, the same limit that
+    already governs forward merges.
diff --git a/test/components/preprocessors/test_embedding_based_document_splitter.py b/test/components/preprocessors/test_embedding_based_document_splitter.py
index 50db2a8aa8..968a770e35 100644
--- a/test/components/preprocessors/test_embedding_based_document_splitter.py
+++ b/test/components/preprocessors/test_embedding_based_document_splitter.py
@@ -215,6 +215,36 @@ def test_merge_small_splits_respect_max_length(self):
         # Second split is merged with third split to get above min_length and still beneath max_length
         assert merged[1] == "1234567891234"
 
+    def test_merge_small_splits_merges_short_trailing_split(self):
+        mock_embedder = Mock()
+        splitter = EmbeddingBasedDocumentSplitter(document_embedder=mock_embedder, min_length=10)
+
+        # The loop only merges forward, so the final accumulator has nothing left to absorb.
+        splits = ["Long enough text ", "Ok."]
+        merged = splitter._merge_small_splits(splits=splits)
+
+        assert merged == ["Long enough text Ok."]
+
+    def test_merge_small_splits_keeps_short_trailing_split_when_max_length_blocks(self):
+        mock_embedder = Mock()
+        splitter = EmbeddingBasedDocumentSplitter(document_embedder=mock_embedder, min_length=10, max_length=15)
+
+        # Merging backwards would reach max_length, so the short tail stays on its own,
+        # matching how a blocked forward merge already behaves.
+        splits = ["123456789012", "1234"]
+        merged = splitter._merge_small_splits(splits=splits)
+
+        assert merged == ["123456789012", "1234"]
+
+    def test_merge_small_splits_keeps_a_lone_short_split(self):
+        mock_embedder = Mock()
+        splitter = EmbeddingBasedDocumentSplitter(document_embedder=mock_embedder, min_length=10)
+
+        # Nothing to merge into.
+        merged = splitter._merge_small_splits(splits=["Ok."])
+
+        assert merged == ["Ok."]
+
     def test_create_documents_from_splits(self):
         mock_embedder = Mock()
         splitter = EmbeddingBasedDocumentSplitter(document_embedder=mock_embedder)