diff --git a/src/lerobot/datasets/streaming_dataset.py b/src/lerobot/datasets/streaming_dataset.py
index 454389d46f..9c37bc9ca7 100644
--- a/src/lerobot/datasets/streaming_dataset.py
+++ b/src/lerobot/datasets/streaming_dataset.py
@@ -94,6 +94,7 @@ def __init__(
         seed: int = 42,
         rng: np.random.Generator | None = None,
         shuffle: bool = True,
+        data_files: str | None = None,
     ):
         """Initialize a StreamingLeRobotDataset.
 
@@ -112,6 +113,8 @@ def __init__(
             seed (int, optional): Reproducibility random seed.
             rng (np.random.Generator | None, optional): Random number generator.
             shuffle (bool, optional): Whether to shuffle the dataset across exhaustions. Defaults to True.
+            data_files (str | None, optional): Pattern to match data files. If None, auto-detects parquet files.
+                Defaults to None.
         """
         super().__init__()
         self.repo_id = repo_id
@@ -128,6 +131,7 @@ def __init__(
 
         self.streaming = streaming
         self.buffer_size = buffer_size
+        self.data_files = data_files
 
         # We cache the video decoders to avoid re-initializing them at each frame (avoiding a ~10x slowdown)
         self.video_decoder_cache = None
@@ -153,7 +157,7 @@ def __init__(
             self.repo_id if not self.streaming_from_local else str(self.root),
             split="train",
             streaming=self.streaming,
-            data_files="data/*/*.parquet",
+            data_files=self.data_files if self.data_files is not None else None,
             revision=self.revision,
         )