diff --git a/python/python/lance/indices/builder.py b/python/python/lance/indices/builder.py
index ca033780a0e..7105c9234bb 100644
--- a/python/python/lance/indices/builder.py
+++ b/python/python/lance/indices/builder.py
@@ -51,7 +51,8 @@ def __init__(self, dataset, column: str):
             the dataset containing the data
         column: str
             The vector column to index, must be a fixed size list of floats
-            or 1-dimensional fixed-shape tensor column.
+            (or unsigned integers for hamming distance) or 1-dimensional
+            fixed-shape tensor column.
         """
         self.dataset = dataset
         self.column = self._normalize_column(column)
@@ -89,7 +90,7 @@ def train_ivf(
             overtraining, reduced recall, and require large nprobes values.  If not
             specified the default will be the integer nearest the square root of the
             number of rows.
-        distance_type: "l2" | "dot" | "cosine"
+        distance_type: "l2" | "dot" | "cosine" | "hamming"
             The distance type to used.  This is defined in more detail in the LanceDB
             documentation on creating indices.
         accelerator: str | torch.Device
@@ -529,6 +530,7 @@ def _normalize_distance_type(self, distance_type):
             "cosine",
             "euclidean",
             "dot",
+            "hamming",
         ]:
             raise ValueError(f"Distance type {distance_type} not supported.")
         return distance_type.lower()
@@ -555,10 +557,13 @@ def _normalize_column(self, column):
                     f"Vector column {c} must be FixedSizeListArray "
                     f"1-dimensional FixedShapeTensorArray, got {field.type}"
                 )
-            if not pa.types.is_floating(field.type.value_type):
+            if not (
+                pa.types.is_floating(field.type.value_type)
+                or pa.types.is_unsigned_integer(field.type.value_type)
+            ):
                 raise TypeError(
-                    f"Vector column {c} must have floating value type, "
-                    f"got {field.type.value_type}"
+                    f"Vector column {c} must have floating or unsigned integer "
+                    f"value type, got {field.type.value_type}"
                 )
 
         return column
diff --git a/python/python/tests/test_indices.py b/python/python/tests/test_indices.py
index e29f02705e2..f0444e4abcb 100644
--- a/python/python/tests/test_indices.py
+++ b/python/python/tests/test_indices.py
@@ -77,6 +77,31 @@ def test_ivf_centroids(tmpdir, rand_dataset):
     assert ivf.centroids == reloaded.centroids
 
 
+def test_ivf_centroids_hamming(tmpdir):
+    num_rows = NUM_ROWS
+    vectors = np.random.randint(0, 256, size=(num_rows, DIMENSION), dtype=np.uint8)
+    vectors_flat = vectors.reshape(-1)
+    vectors_arr = pa.FixedSizeListArray.from_arrays(
+        pa.array(vectors_flat, type=pa.uint8()), DIMENSION
+    )
+    table = pa.Table.from_arrays([vectors_arr], names=["vectors"])
+    uri = str(tmpdir / "hamming_dataset")
+    ds = lance.write_dataset(table, uri, max_rows_per_file=NUM_ROWS_PER_FRAGMENT)
+
+    ivf = IndicesBuilder(ds, "vectors").train_ivf(
+        sample_rate=16, distance_type="hamming"
+    )
+
+    assert ivf.distance_type == "hamming"
+    expected_partitions = round(math.sqrt(num_rows))
+    assert len(ivf.centroids) == expected_partitions
+
+    ivf.save(str(tmpdir / "ivf_hamming"))
+    reloaded = IvfModel.load(str(tmpdir / "ivf_hamming"))
+    assert reloaded.distance_type == "hamming"
+    assert ivf.centroids == reloaded.centroids
+
+
 @pytest.mark.parametrize("distance_type", ["l2", "cosine", "dot"])
 def test_ivf_centroids_mostly_null(mostly_null_dataset, distance_type):
     ivf = IndicesBuilder(mostly_null_dataset, "vectors").train_ivf(