diff --git a/FlagEmbedding/abc/evaluation/utils.py b/FlagEmbedding/abc/evaluation/utils.py
index 9ed3545dc..6583c22d8 100644
--- a/FlagEmbedding/abc/evaluation/utils.py
+++ b/FlagEmbedding/abc/evaluation/utils.py
@@ -31,8 +31,12 @@ def evaluate_mrr(
     k_max, top_hits = max(k_values), {}
 
     for query_id, doc_scores in results.items():
+        # A score-only sort keeps dict order, so a tied relevant document can
+        # rank behind an earlier id. Break ties by document id, higher first.
         top_hits[query_id] = sorted(
-            doc_scores.items(), key=lambda item: item[1], reverse=True
+            doc_scores.items(),
+            key=lambda item: (item[1], item[0]),
+            reverse=True,
         )[0:k_max]
 
     for query_id in top_hits:
@@ -77,7 +81,11 @@ def evaluate_recall_cap(
     logging.info("\n")
     
     for query_id, doc_scores in results.items():
-        top_hits = sorted(doc_scores.items(), key=lambda item: item[1], reverse=True)[0:k_max]   
+        top_hits = sorted(
+            doc_scores.items(),
+            key=lambda item: (item[1], item[0]),
+            reverse=True,
+        )[0:k_max]   
         query_relevant_docs = [doc_id for doc_id in qrels[query_id] if qrels[query_id][doc_id] > 0]
         for k in k_values:
             retrieved_docs = [row[0] for row in top_hits[0:k] if qrels[query_id].get(row[0], 0) > 0]
diff --git a/tests/test_evaluate_mrr.py b/tests/test_evaluate_mrr.py
new file mode 100644
index 000000000..3bd662dd9
--- /dev/null
+++ b/tests/test_evaluate_mrr.py
@@ -0,0 +1,20 @@
+from FlagEmbedding.abc.evaluation.utils import evaluate_mrr, evaluate_recall_cap
+
+
+def test_a_tied_relevant_doc_is_ranked_by_id():
+    qrels = {"q": {"b": 1, "a": 0}}
+    results = {"q": {"a": 1.0, "b": 1.0}}
+
+    mrr = evaluate_mrr(qrels, results, [1])
+    assert mrr["MRR@1"] == 1.0
+
+    capped = evaluate_recall_cap(qrels, results, [1])
+    assert capped["R_cap@1"] == 1.0
+
+
+def test_a_higher_score_still_ranks_first():
+    qrels = {"q": {"b": 1}}
+    results = {"q": {"a": 0.2, "b": 0.9}}
+
+    mrr = evaluate_mrr(qrels, results, [1])
+    assert mrr["MRR@1"] == 1.0