diff --git a/config/config.flores.yml b/config/config.flores.yml
index d5260d3..bfaa831 100644
--- a/config/config.flores.yml
+++ b/config/config.flores.yml
@@ -64,6 +64,8 @@ scorer:
       metric_label: tok_len_var_bytes
       mode: var
       use_bytes: true
+    - metric: eflomal
+      metric_label: eflomal_score
 
   languages_info: "tokcollate/resources/language/languages.json"
   system_dataset_suffix: "txt"
diff --git a/requirements-all.txt b/requirements-all.txt
index b3c73b2..079c0aa 100644
--- a/requirements-all.txt
+++ b/requirements-all.txt
@@ -6,3 +6,4 @@ pytest==8.3.5
 sentencepiece
 scipy
 typing-extensions
+eflomal
\ No newline at end of file
diff --git a/requirements.txt b/requirements.txt
index 4e26ad0..41d3ffe 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -6,3 +6,4 @@ omegaconf
 sentencepiece
 scipy
 typing-extensions
+eflomal
diff --git a/tokcollate/metrics/eflomal.py b/tokcollate/metrics/eflomal.py
new file mode 100644
index 0000000..d030a06
--- /dev/null
+++ b/tokcollate/metrics/eflomal.py
@@ -0,0 +1,85 @@
+
+import tempfile
+from io import StringIO
+
+import numpy as np
+from attr import define
+
+from tokcollate.data import TokCollateData
+from tokcollate.metrics import register_metric
+from tokcollate.metrics.tokcollate_metric import TokCollateMultilingualMetric
+
+import eflomal
+
+@register_metric("eflomal")
+@define(kw_only=True)
+class EflomalScore(TokCollateMultilingualMetric):
+    """Measures the Eflomal Score over the (parallel) text vocabulary distributions.
+
+    This metric uses the Eflomal aligner to compute alignment scores between the
+    source and target tokens. The final score is computed as the average of the
+    forward and reverse alignment scores of the alignment between the source and
+    target tokens.
+
+    See https://aclanthology.org/2025.naacl-short.63 for more details.    
+    """
+
+    def score(self, data: TokCollateData, system_label: str, src_lang: str, tgt_lang: str) -> float:
+        text_src = data.get_system_text(system_label=system_label, language=src_lang)
+        text_tgt = data.get_system_text(system_label=system_label, language=tgt_lang)
+
+        aligner = eflomal.Aligner()
+
+        # Create file-like objects from the text variables
+        src_data = StringIO("\n".join([" ".join(line) for line in text_src]))
+        trg_data = StringIO("\n".join([" ".join(line) for line in text_tgt]))
+        
+        aligner = eflomal.Aligner()
+
+        # Create temporary files for alignment scores
+        with tempfile.NamedTemporaryFile(mode='w', suffix='.fwd.scores', delete=False) as fwd_scores_file, \
+                tempfile.NamedTemporaryFile(mode='w', suffix='.rev.scores', delete=False) as rev_scores_file:
+            fwd_scores_path = fwd_scores_file.name
+            rev_scores_path = rev_scores_file.name
+
+        try:
+            aligner.align(
+                src_data, trg_data,
+                scores_filename_fwd=fwd_scores_path,
+                scores_filename_rev=rev_scores_path)
+            
+            # Read the alignment scores
+            with open(fwd_scores_path, 'r', encoding='utf-8') as f:
+                fwd_scores = f.read()
+            with open(rev_scores_path, 'r', encoding='utf-8') as f:
+                rev_scores = f.read()
+
+        finally:
+            # Clean up temporary files
+            import os
+            if os.path.exists(fwd_scores_path):
+                os.unlink(fwd_scores_path)
+            if os.path.exists(rev_scores_path):
+                os.unlink(rev_scores_path)
+
+        fwd_scores_list = [float(score) for score in fwd_scores.splitlines()]
+        rev_scores_list = [float(score) for score in rev_scores.splitlines()]
+
+        # Filter out inf values and compute mean (nanmean handles NaN values)
+        fwd_scores_filtered = [s for s in fwd_scores_list if not np.isinf(s)]
+        rev_scores_filtered = [s for s in rev_scores_list if not np.isinf(s)]
+
+        fwd_mean = np.nanmean(fwd_scores_filtered) if fwd_scores_filtered else np.nan
+        rev_mean = np.nanmean(rev_scores_filtered) if rev_scores_filtered else np.nan
+
+        return (fwd_mean + rev_mean) / 2
+
+    def score_batched(self, data: TokCollateData, system_label: str, languages: list[str]) -> np.ndarray:
+        result = np.zeros((len(languages), len(languages)))
+        for i, src_lang in enumerate(languages):
+            for j, tgt_lang in enumerate(languages):
+                if i <= j:
+                    score = self.score(data, system_label, src_lang, tgt_lang)
+                    result[i, j] = score
+                    result[j, i] = score  # Symmetric score
+        return result
\ No newline at end of file