diff --git a/src/setfit/model_card.py b/src/setfit/model_card.py
index 30706bcc..1cdebbf4 100644
--- a/src/setfit/model_card.py
+++ b/src/setfit/model_card.py
@@ -4,6 +4,7 @@
 from dataclasses import dataclass, field, fields
 from pathlib import Path
 from platform import python_version
+from statistics import median
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union
 
 import datasets
@@ -312,7 +313,7 @@ def add_naive_word_count(sample: Dict[str, Any]) -> Dict[str, Any]:
             {
                 "Training set": "Word count",
                 "Min": min(dataset["word_count"]),
-                "Median": sum(dataset["word_count"]) / len(dataset),
+                "Median": median(dataset["word_count"]),
                 "Max": max(dataset["word_count"]),
             },
         ]
diff --git a/tests/model_card_pattern.py b/tests/model_card_pattern.py
index 59e290ec..9eaf4b52 100644
--- a/tests/model_card_pattern.py
+++ b/tests/model_card_pattern.py
@@ -99,7 +99,7 @@
 ### Training Set Metrics
 \| Training set \| Min \| Median \| Max \|
 \|:-------------\|:----\|:-------\|:----\|
-\| Word count   \| 3   \| 7.875  \| 18  \|
+\| Word count   \| 3   \| [\d.]+ +\| 18  \|
 
 \| Label    \| Training Sample Count \|
 \|:---------\|:----------------------\|
diff --git a/tests/span/aspect_model_card_pattern.py b/tests/span/aspect_model_card_pattern.py
index 7cf7393b..24cd46a2 100644
--- a/tests/span/aspect_model_card_pattern.py
+++ b/tests/span/aspect_model_card_pattern.py
@@ -111,7 +111,7 @@
 ### Training Set Metrics
 \| Training set \| Min \| Median \| Max \|
 \|:-------------\|:----\|:-------\|:----\|
-\| Word count   \| 5   \| 14.5   \| 23  \|
+\| Word count   \| 5   \| [\d.]+ +\| 23  \|
 
 \| Label     \| Training Sample Count \|
 \|:----------\|:----------------------\|
diff --git a/tests/span/polarity_model_card_pattern.py b/tests/span/polarity_model_card_pattern.py
index e9eabf67..5e375170 100644
--- a/tests/span/polarity_model_card_pattern.py
+++ b/tests/span/polarity_model_card_pattern.py
@@ -111,7 +111,7 @@
 ### Training Set Metrics
 \| Training set \| Min \| Median \| Max \|
 \|:-------------\|:----\|:-------\|:----\|
-\| Word count   \| 8   \| 16.8   \| 28  \|
+\| Word count   \| 8   \| [\d.]+ +\| 28  \|
 
 \| Label    \| Training Sample Count \|
 \|:---------\|:----------------------\|
diff --git a/tests/test_model_card.py b/tests/test_model_card.py
index 9bdb199c..8aab2e8b 100644
--- a/tests/test_model_card.py
+++ b/tests/test_model_card.py
@@ -62,6 +62,26 @@ def test_model_card_languages() -> None:
     assert "**Languages:** en, nl, de" in model_card
 
 
+def test_train_set_metrics_word_count() -> None:
+    # The word counts are 2, 3, 4 and 11, i.e. a median of 3.5 and a mean of 5.0
+    train_dataset = Dataset.from_dict(
+        {
+            "text": [
+                "one two",
+                "one two three",
+                "one two three four",
+                "one two three four five six seven eight nine ten eleven",
+            ],
+            "label": [0, 1, 0, 1],
+        }
+    )
+    model_card_data = SetFitModelCardData()
+    model_card_data.set_train_set_metrics(train_dataset)
+    assert model_card_data.train_set_metrics_list == [
+        {"Training set": "Word count", "Min": 2, "Median": 3.5, "Max": 11}
+    ]
+
+
 def test_is_on_huggingface_edge_case() -> None:
     assert not is_on_huggingface("test_value")
     assert not is_on_huggingface("a/test/value")