diff --git a/mmfewshot/classification/core/evaluation/eval_hooks.py b/mmfewshot/classification/core/evaluation/eval_hooks.py
index d324025..e088be5 100644
--- a/mmfewshot/classification/core/evaluation/eval_hooks.py
+++ b/mmfewshot/classification/core/evaluation/eval_hooks.py
@@ -86,8 +86,8 @@ def before_run(self, runner: Runner) -> None:
                 warnings.warn('runner.meta is None. Creating an empty one.')
                 runner.meta = dict()
             runner.meta.setdefault('hook_msgs', dict())
-            self.best_score = runner.meta.get('best_score', 0.0)
-            if self.best_score > 0.0:
+            if runner.meta['hook_msgs'].get('best_score', False):
+                self.best_score = runner.meta['hook_msgs']['best_score']
                 runner.logger.info(
                     f'Previous best score is: {self.best_score}.')
             self.best_ckpt_path = runner.meta['hook_msgs'].get(
@@ -140,7 +140,6 @@ def _save_ckpt(self, runner: Runner, key_score: float) -> None:
 
         if self.best_score < key_score:
             self.best_score = key_score
-            runner.meta['best_score'] = self.best_score
             runner.meta['hook_msgs']['best_score'] = self.best_score
             runner.meta['hook_msgs']['ckpt_time'] = current
 
@@ -152,10 +151,7 @@ def _save_ckpt(self, runner: Runner, key_score: float) -> None:
             runner.meta['hook_msgs']['best_ckpt'] = self.best_ckpt_path
 
             runner.save_checkpoint(
-                runner.work_dir,
-                best_ckpt_name,
-                meta={'best_score': self.best_score},
-                create_symlink=False)
+                runner.work_dir, best_ckpt_name, create_symlink=False)
             runner.logger.info(
                 f'Now best checkpoint is saved as {best_ckpt_name}.')
             runner.logger.info(
diff --git a/tests/test_classification_runtime/test_meta_test_eval_hook.py b/tests/test_classification_runtime/test_meta_test_eval_hook.py
index 2b3c626..c2c7039 100644
--- a/tests/test_classification_runtime/test_meta_test_eval_hook.py
+++ b/tests/test_classification_runtime/test_meta_test_eval_hook.py
@@ -213,6 +213,6 @@ def test_resume_eval_hook():
             logger=logging.getLogger(),
             max_epochs=1)
         runner.register_hook(eval_hook)
-        runner.meta = {'best_score': 99.0}
+        runner.meta = {'hook_msgs': {'best_score': 99.0}}
         runner.run([loader], [('train', 1)], 1)
         assert eval_hook.best_score == 99.0