diff --git a/nemo/collections/asr/parts/utils/eou_utils.py b/nemo/collections/asr/parts/utils/eou_utils.py
index 256e8b418951..fb81aeb19e0f 100644
--- a/nemo/collections/asr/parts/utils/eou_utils.py
+++ b/nemo/collections/asr/parts/utils/eou_utils.py
@@ -173,8 +173,12 @@ def evaluate_eou(
         false_negatives += len(reference) - r_idx
         missing += len(reference) - r_idx
 
-    missing -= len(earlycut_ids)  # Remove the references that were missed due to early cutoff
-    false_negatives -= len(earlycut_ids)  # Remove the references that were missed due to early cutoff
+    # Only references counted by the block above can be discounted here. An early cutoff
+    # that r_idx already advanced past was never added, so subtracting it drives both
+    # counts negative.
+    counted_earlycut = sum(1 for idx in earlycut_ids if idx >= r_idx)
+    missing -= counted_earlycut  # Remove the references that were missed due to early cutoff
+    false_negatives -= counted_earlycut  # Remove the references that were missed due to early cutoff
     return EOUResult(
         latency=latency,
         early_cutoff=early_cutoff,
diff --git a/tests/collections/asr/test_asr_eou.py b/tests/collections/asr/test_asr_eou.py
index 5446f2ac5eb5..927118b88674 100644
--- a/tests/collections/asr/test_asr_eou.py
+++ b/tests/collections/asr/test_asr_eou.py
@@ -185,3 +185,32 @@ def test_cal_eou_metrics_from_frame_labels_multiple_eou(self):
         assert eou_metrics.missing == 0
         assert eou_metrics.early_cutoff == []
         assert np.allclose(eou_metrics.latency, [delay] * len(ref_eou_times))
+
+
+class TestEvaluateEOUCounts:
+    def test_counts_stay_non_negative_when_early_cutoff_is_skipped(self):
+        """An early cutoff that r_idx advances past is never counted, so it must not be subtracted."""
+        reference = [
+            {"start_time": 0.0, "end_time": 2.0},
+            {"start_time": 5.0, "end_time": 7.0},
+        ]
+        prediction = [
+            {"start_time": 0.0, "end_time": 1.0, "eou_prob": 0.9},
+            {"start_time": 5.5, "end_time": 7.0, "eou_prob": 0.9},
+        ]
+
+        result = evaluate_eou(prediction=prediction, reference=reference, threshold=None, collar=0.1)
+
+        assert result.false_negatives == 0
+        assert result.missing == 0
+
+    def test_trailing_early_cutoff_is_still_discounted(self):
+        """A trailing early cutoff is counted by the tail, so the discount must still apply."""
+        reference = [{"start_time": 0.0, "end_time": 2.0}]
+        prediction = [{"start_time": 0.0, "end_time": 1.0, "eou_prob": 0.9}]
+
+        result = evaluate_eou(prediction=prediction, reference=reference, threshold=None, collar=0.1)
+
+        assert result.false_negatives == 0
+        assert result.missing == 0
+        assert len(result.early_cutoff) == 1