diff --git a/rest-api-spec/src/main/resources/rest-api-spec/test/search/50_multi_match.yml b/rest-api-spec/src/main/resources/rest-api-spec/test/search/50_multi_match.yml
new file mode 100644
index 0000000000000..b20438c45b33d
--- /dev/null
+++ b/rest-api-spec/src/main/resources/rest-api-spec/test/search/50_multi_match.yml
@@ -0,0 +1,32 @@
+"Cross fields do not return negative scores":
+  - do:
+      index:
+        index: test
+        id: 1
+        body: { "color" : "orange red yellow" }
+  - do:
+      index:
+        index: test
+        id: 2
+        body: { "color": "orange red purple", "shape": "red square" }
+  - do:
+      index:
+        index: test
+        id: 3
+        body: { "color" : "orange red yellow purple" }
+  - do:
+      indices.refresh: { }
+  - do:
+      search:
+        index: test
+        body:
+          query:
+            multi_match:
+              query: "red"
+              type: "cross_fields"
+              fields: [ "color", "shape^100"]
+              tie_breaker: 0.1
+          explain: true
+  - match: { hits.total.value: 3 }
+  - match: { hits.hits.0._id: "2" }
+  - gt: { hits.hits.2._score: 0.0 }
diff --git a/server/src/main/java/org/apache/lucene/queries/BlendedTermQuery.java b/server/src/main/java/org/apache/lucene/queries/BlendedTermQuery.java
index b47b974b96fed..17419804b1ffa 100644
--- a/server/src/main/java/org/apache/lucene/queries/BlendedTermQuery.java
+++ b/server/src/main/java/org/apache/lucene/queries/BlendedTermQuery.java
@@ -120,6 +120,7 @@ protected void blend(final TermStates[] contexts, int maxDoc, IndexReader reader
         }
         int max = 0;
         long minSumTTF = Long.MAX_VALUE;
+        int minDocCount = Integer.MAX_VALUE;
         for (int i = 0; i < contexts.length; i++) {
             TermStates ctx = contexts[i];
             int df = ctx.docFreq();
@@ -133,11 +134,15 @@ protected void blend(final TermStates[] contexts, int maxDoc, IndexReader reader
                 // we need to find out the minimum sumTTF to adjust the statistics
                 // otherwise the statistics don't match
                 minSumTTF = Math.min(minSumTTF, reader.getSumTotalTermFreq(terms[i].field()));
+                minDocCount = Math.min(minDocCount, reader.getDocCount(terms[i].field()));
             }
         }
         if (maxDoc > minSumTTF) {
             maxDoc = (int) minSumTTF;
         }
+        if (maxDoc > minDocCount) {
+            maxDoc = minDocCount;
+        }
         if (max == 0) {
             return; // we are done that term doesn't exist at all
         }