diff --git a/tests/test_sgkit.py b/tests/test_sgkit.py
index b930263a..8167474a 100644
--- a/tests/test_sgkit.py
+++ b/tests/test_sgkit.py
@@ -600,6 +600,22 @@ def test_ploidy1_unphased(self, tmp_path):
         sgkit.save_dataset(ds, path)
         tsinfer.SgkitSampleData(path)
 
+    def test_duplicate_positions(self, tmp_path):
+        path = tmp_path / "data.zarr"
+        ds = sgkit.simulate_genotype_call_dataset(n_variant=3, n_sample=3, phased=True)
+        ds["variant_position"][2] = ds["variant_position"][1]
+        sgkit.save_dataset(ds, path)
+        with pytest.raises(ValueError, match="duplicate or out-of-order values"):
+            tsinfer.SgkitSampleData(path)
+
+    def test_bad_order_positions(self, tmp_path):
+        path = tmp_path / "data.zarr"
+        ds = sgkit.simulate_genotype_call_dataset(n_variant=3, n_sample=3, phased=True)
+        ds["variant_position"][0] = ds["variant_position"][2] - 0.5
+        sgkit.save_dataset(ds, path)
+        with pytest.raises(ValueError, match="duplicate or out-of-order values"):
+            tsinfer.SgkitSampleData(path)
+
     def test_empty_alleles_not_at_end(self, tmp_path):
         path = tmp_path / "data.zarr"
         ds = sgkit.simulate_genotype_call_dataset(n_variant=3, n_sample=3, n_ploidy=1)
diff --git a/tsinfer/formats.py b/tsinfer/formats.py
index 5ae575b7..f578f435 100644
--- a/tsinfer/formats.py
+++ b/tsinfer/formats.py
@@ -2309,6 +2309,12 @@ def __init__(self, path):
                     " sgkit dataset, indicating that all the genotypes are"
                     " unphased"
                 )
+        if np.any(np.diff(self.sites_position) <= 0):
+            raise ValueError(
+                "Values taken from the variant_position array are not strictly "
+                "increasing (i.e. have duplicate or out-of-order values). "
+                "These must be masked out to run tsinfer."
+            )
 
     @functools.cached_property
     def format_name(self):