diff --git a/mrQA/tests/conftest.py b/mrQA/tests/conftest.py
index 3772bc2..3a10141 100644
--- a/mrQA/tests/conftest.py
+++ b/mrQA/tests/conftest.py
@@ -1,3 +1,4 @@
+import tempfile
 import typing as tp
 from pathlib import Path
 from typing import Tuple
@@ -42,9 +43,11 @@ def create_dataset(draw_from: st.DrawFn) -> Tuple:
                                               repetition_time,
                                               echo_train_length,
                                               flip_angle)
+    temp_dir = Path(tempfile.mkdtemp())
     ds = DicomDataset(name=name,
                       data_source=fake_ds_dir,
-                      config_path=THIS_DIR / 'resources/mri-config.json')
+                      config_path=THIS_DIR / 'resources/mri-config.json',
+                      output_dir=temp_dir)
     ref_protocol_path = sample_protocol()
     attributes = {
         'name': name,
diff --git a/mrQA/tests/test_parallel.py b/mrQA/tests/test_parallel.py
index 52bbb67..01de999 100644
--- a/mrQA/tests/test_parallel.py
+++ b/mrQA/tests/test_parallel.py
@@ -29,7 +29,8 @@ def test_equivalence_seq_vs_parallel():
             sequential_ds = import_dataset(data_source=data_source,
                                            ds_format='dicom',
                                            name='sequential',
-                                           config_path=config_path)
+                                           config_path=config_path,
+                                           output_dir=output_dir)
             save_mr_dataset(output_path['sequential'], sequential_ds)
         else:
             sequential_ds = load_mr_dataset(output_path['sequential'])
@@ -83,58 +84,61 @@ def test_equivalence_seq_vs_parallel():
 def test_merging():
     # Sequential complete processing of the dataset
     data_source = sample_dicom_dataset()
-    output_dir = Path(data_source).parent / 'test_merge_mrqa_files'
-    output_path_seq = output_dir / ('sequential' + MRDS_EXT)
-    config_path = THIS_DIR / 'resources/mri-config.json'
-
-    if not output_path_seq.exists():
-        sequential_ds = import_dataset(data_source=data_source,
-                                       ds_format='dicom',
-                                       name='sequential',
-                                       config_path=config_path)
-        save_mr_dataset(output_path_seq, sequential_ds)
-    else:
-        sequential_ds = load_mr_dataset(output_path_seq)
-
-    # Start processing in batches
-    folder_paths, files_per_batch, all_ids_path = _make_file_folders(output_dir)
-
-    # For each batch create the list of ids to be processed
-    ids_path_list = split_folders_list(
-        data_source,
-        all_fnames_path=all_ids_path,
-        per_batch_ids=files_per_batch['fnames'],
-        output_dir=folder_paths['fnames'],
-        folders_per_job=5
-    )
-
-    # The paths to the output files
-    output_path = {i: output_dir/f'seq{i}{MRDS_EXT}'
-                   for i in range(len(ids_path_list))}
-    ds_list = []
-    for i, filepath in enumerate(ids_path_list):
-        # Read the list of subject ids to be processed
-        subject_folders_list = txt2list(filepath)
-        if not output_path[i].exists():
-            # Process the batch of subjects
-            ds = import_dataset(data_source=subject_folders_list,
-                                ds_format='dicom',
-                                name=f'seq{i}',
-                                config_path=config_path)
-            save_mr_dataset(output_path[i], ds)
-        else:
-            ds = load_mr_dataset(output_path[i])
-        ds_list.append(ds)
-
-    # Merge batches
-    combined_mrds = None
-    for ds in ds_list:
-        if combined_mrds is None:
-            # Add the first partial dataset
-            combined_mrds = ds
-        else:
-            # otherwise, keep aggregating
-            combined_mrds.merge(ds)
+    with tempfile.TemporaryDirectory() as tempdir:
+        output_dir = Path(tempdir)
+        output_path_seq = output_dir / ('sequential' + MRDS_EXT)
+        config_path = THIS_DIR / 'resources/mri-config.json'
 
-    # Check if both datasets are the same
-    assert combined_mrds == sequential_ds
+        if not output_path_seq.exists():
+            sequential_ds = import_dataset(data_source=data_source,
+                                           ds_format='dicom',
+                                           name='sequential',
+                                           config_path=config_path,
+                                           output_dir=output_dir)
+            save_mr_dataset(output_path_seq, sequential_ds)
+        else:
+            sequential_ds = load_mr_dataset(output_path_seq)
+
+        # Start processing in batches
+        folder_paths, files_per_batch, all_ids_path = _make_file_folders(output_dir)
+
+        # For each batch create the list of ids to be processed
+        ids_path_list = split_folders_list(
+            data_source,
+            all_fnames_path=all_ids_path,
+            per_batch_ids=files_per_batch['fnames'],
+            output_dir=folder_paths['fnames'],
+            folders_per_job=5
+        )
+
+        # The paths to the output files
+        output_path = {i: folder_paths['mrds']/f'seq{i}{MRDS_EXT}'
+                       for i in range(len(ids_path_list))}
+        ds_list = []
+        for i, filepath in enumerate(ids_path_list):
+            # Read the list of subject ids to be processed
+            subject_folders_list = txt2list(filepath)
+            if not output_path[i].exists():
+                # Process the batch of subjects
+                ds = import_dataset(data_source=subject_folders_list,
+                                    ds_format='dicom',
+                                    name=f'seq{i}',
+                                    config_path=config_path,
+                                    output_dir=folder_paths['mrds'])
+                save_mr_dataset(output_path[i], ds)
+            else:
+                ds = load_mr_dataset(output_path[i])
+            ds_list.append(ds)
+
+        # Merge batches
+        combined_mrds = None
+        for ds in ds_list:
+            if combined_mrds is None:
+                # Add the first partial dataset
+                combined_mrds = ds
+            else:
+                # otherwise, keep aggregating
+                combined_mrds.merge(ds)
+        save_mr_dataset(output_dir / ('parallel' + MRDS_EXT), combined_mrds)
+        # Check if both datasets are the same
+        assert combined_mrds == sequential_ds