diff --git a/hands_on/pyanno_voting/pyanno/tests/test_voting.py b/hands_on/pyanno_voting/pyanno/tests/test_voting.py
index f21a10c..2cbf4de 100644
--- a/hands_on/pyanno_voting/pyanno/tests/test_voting.py
+++ b/hands_on/pyanno_voting/pyanno/tests/test_voting.py
@@ -41,3 +41,10 @@ def test_majority_vote_empty_item():
     expected = [1, MV, 2]
     result = voting.majority_vote(annotations)
     assert result == expected
+
+
+def test_labels_frequency():
+    result = voting.labels_frequency([[1, 1, 2], [MV, 1, 2]], 4)
+    expected = np.array([ 0. ,  0.6,  0.4,  0. ])
+    np.testing.assert_array_equal(result, expected)
+
diff --git a/hands_on/pyanno_voting/pyanno/voting.py b/hands_on/pyanno_voting/pyanno/voting.py
index d5b5747..a461e08 100644
--- a/hands_on/pyanno_voting/pyanno/voting.py
+++ b/hands_on/pyanno_voting/pyanno/voting.py
@@ -100,3 +100,21 @@ def labels_frequency(annotations, nclasses):
         freq[k] is the frequency of elements of class k in `annotations`, i.e.
         their count over the number of total of observed (non-missing) elements
     """
+    
+    annotations = np.asarray(annotations)
+    nitems = annotations.shape[0]
+    valid = annotations != MISSING_VALUE
+
+    annotations_without_mv = annotations[valid]
+
+    unique, counts = np.unique(annotations_without_mv, return_counts=True)
+
+    output = np.zeros(nclasses)
+    j = 0
+    for i in range(nclasses):
+        if i in unique:
+            output[i] = counts[j] / sum(counts)
+            j += 1
+            
+    return(output)
+