diff --git a/openevolve/database.py b/openevolve/database.py
index ab1dad21b..0551e6c89 100644
--- a/openevolve/database.py
+++ b/openevolve/database.py
@@ -281,7 +281,8 @@ def add(
             target_island: Specific island to add to (auto-detects parent's island if None)
 
         Returns:
-            Program ID, or None when the admission strategy rejects the candidate
+            Program ID, or None when the admission strategy or the novelty check
+            rejects the candidate
         """
         if self.population_strategy is None or not any(
             (
@@ -382,17 +383,25 @@ def _add(
         if iteration is not None:
             program.iteration_found = iteration
             self.last_iteration = max(self.last_iteration, iteration)
+        previous_entry = self.programs.get(program.id)
         self.programs[program.id] = program
 
-        # Calculate feature coordinates for MAP-Elites
-        feature_coords = self._calculate_feature_coords(program)
-
         # Novelty check before adding
         if not self._is_novel(program.id, island_idx):
             logger.debug(
                 f"Program {program.id} failed in novelty check and won't be added in the island {island_idx}"
             )
-            return program.id  # Do not add non-novel program
+            # Undo the provisional store so the rejected program cannot be sampled,
+            # counted against the population limit, or saved to checkpoints.
+            if previous_entry is None:
+                del self.programs[program.id]
+            else:
+                self.programs[program.id] = previous_entry
+            return None
+
+        # Calculate feature coordinates for MAP-Elites (after the novelty check so
+        # rejected programs do not widen the feature scaling ranges)
+        feature_coords = self._calculate_feature_coords(program)
 
         # Add to island-specific feature map (replacing existing if better)
         feature_key = self._feature_coords_to_key(feature_coords)
diff --git a/openevolve/process_parallel.py b/openevolve/process_parallel.py
index 8c2b29adf..ee45f8447 100644
--- a/openevolve/process_parallel.py
+++ b/openevolve/process_parallel.py
@@ -959,7 +959,7 @@ async def run_evolution(
                                         break
 
             except _CandidateRejected as rejected:
-                logger.info("Admission strategy rejected program %s", rejected)
+                logger.info("Database rejected program %s", rejected)
                 try:
                     self._checkpoint_if_due(completed_iteration, checkpoint_callback)
                 except Exception as e:
diff --git a/tests/test_novelty_rejection.py b/tests/test_novelty_rejection.py
new file mode 100644
index 000000000..2cbf21787
--- /dev/null
+++ b/tests/test_novelty_rejection.py
@@ -0,0 +1,89 @@
+"""Tests that programs rejected by the novelty check are not kept in the database."""
+
+import tempfile
+import unittest
+from unittest.mock import MagicMock, patch
+
+from openevolve.config import Config
+from openevolve.database import Program, ProgramDatabase
+
+
+class NotNovelLLM:
+    async def generate_with_context(self, system_message: str, messages: list) -> str:
+        return "NOT_NOVEL"
+
+
+class TestNoveltyRejection(unittest.TestCase):
+    @patch("openevolve.embedding.EmbeddingClient")
+    def setUp(self, mock_embedding_client_class) -> None:
+        # Identical embeddings so every candidate is sent to the (rejecting) LLM judge
+        mock_embedding_client_class.return_value = MagicMock(
+            get_embedding=MagicMock(return_value=[1.0, 0.0])
+        )
+        config = Config()
+        config.database.in_memory = True
+        config.database.db_path = None
+        config.database.num_islands = 1
+        config.database.population_size = 3
+        config.database.embedding_model = "text-embedding-3-small"
+        config.database.similarity_threshold = 0.5
+        config.database.novelty_llm = NotNovelLLM()
+        self.db = ProgramDatabase(config.database)
+
+        self.seed = Program(id="seed", code="x = 0", metrics={"combined_score": 0.1})
+        self.assertEqual(self.db.add(self.seed), "seed")
+
+    def add_rejected(self, n: int) -> list:
+        return [
+            self.db.add(
+                Program(
+                    id=f"dup{i}",
+                    code=f"x = {i + 1}",
+                    parent_id="seed",
+                    metrics={"combined_score": 0.9},
+                )
+            )
+            for i in range(n)
+        ]
+
+    def test_rejected_program_is_not_stored(self) -> None:
+        results = self.add_rejected(5)
+
+        self.assertEqual(results, [None] * 5)
+        self.assertEqual(set(self.db.programs), {"seed"})
+        self.assertEqual(self.db.islands, [{"seed"}])
+        self.assertLessEqual(len(self.db.programs), self.db.config.population_size)
+
+    def test_rejected_program_is_never_sampled_or_ranked(self) -> None:
+        self.add_rejected(5)
+
+        for _ in range(20):
+            self.assertEqual(self.db._sample_random_parent().id, "seed")
+        self.assertEqual([p.id for p in self.db.get_top_programs(n=10)], ["seed"])
+
+    def test_rejected_program_is_not_checkpointed(self) -> None:
+        self.add_rejected(2)
+
+        with tempfile.TemporaryDirectory() as tmp:
+            self.db.save(tmp)
+            config = Config()
+            config.database.in_memory = True
+            config.database.db_path = None
+            config.database.num_islands = 1
+            restored = ProgramDatabase(config.database)
+            restored.load(tmp)
+
+        self.assertEqual(set(restored.programs), {"seed"})
+
+    def test_rejection_does_not_touch_feature_stats(self) -> None:
+        stats_before = {
+            k: dict(v, values=list(v["values"])) for k, v in self.db.feature_stats.items()
+        }
+
+        self.db.add(Program(id="long", code="y = 1\n" * 500, metrics={"combined_score": 0.9}))
+
+        self.assertEqual(self.db.feature_stats, stats_before)
+
+
+if __name__ == "__main__":
+    unittest.main()