diff --git a/src/annbatch/loader.py b/src/annbatch/loader.py
index 2286c761..d5c85dbb 100644
--- a/src/annbatch/loader.py
+++ b/src/annbatch/loader.py
@@ -998,6 +998,11 @@ def __iter__(
             in_memory_indices: None | np.ndarray = self._maybe_accumulate_indices(dataset_index_to_rows)
             for split in splits:
                 sel = inv[split]
+
+                # Use basic slicing for contiguous selections to avoid costly fancy indexing on the loaded memory
+                if len(sel) > 0 and (sel[-1] - sel[0] == len(sel) - 1 and np.all(np.diff(sel) == 1)):
+                    sel = slice(sel[0], sel[-1] + 1)
+
                 data = in_memory_data[sel]
                 yield {
                     "X": data if not self._to_torch else to_torch(data, self._preload_to_gpu),