diff --git a/config.py b/config.py
index 2a8ee39db..9234255d5 100644
--- a/config.py
+++ b/config.py
@@ -2020,6 +2020,14 @@ def gen_log_kwargs(
     return kwargs
 
 
+###############################################################################
+# Private config vars (not to be set by user)
+# -------------------------------------------
+
+_raw_split_size = '2GB'
+_epochs_split_size = '2GB'
+
+
 ###############################################################################
 # Retrieve custom configuration options
 # -------------------------------------
@@ -3824,3 +3832,25 @@ def save_logs(logs):
            'configuration. Currently the `conditions` parameter is empty. '
            'This is only allowed for resting-state analysis.')
     raise ValueError(msg)
+
+
+def _update_for_splits(files_dict, key, *, single=False):
+    if not isinstance(files_dict, dict):  # fake it
+        assert key is None
+        files_dict, key = dict(x=files_dict), 'x'
+    bids_path = files_dict[key]
+    if bids_path.fpath.exists():
+        return bids_path  # no modifications needed
+    bids_path = bids_path.copy().update(split='01')
+    assert bids_path.fpath.exists(), f'Missing file: {bids_path.fpath}'
+    files_dict[key] = bids_path
+    # if we only need the first file (i.e., when reading), quit now
+    if single:
+        return bids_path
+    for split in range(2, 100):
+        split_key = f'{split:02d}'
+        bids_path_next = bids_path.copy().update(split=split_key)
+        if not bids_path_next.fpath.exists():
+            break
+        files_dict[f'{key}_split-{split_key}'] = bids_path_next
+    return bids_path
diff --git a/docs/source/changes.md b/docs/source/changes.md
index b6fa8afd6..da33dc1c3 100644
--- a/docs/source/changes.md
+++ b/docs/source/changes.md
@@ -352,3 +352,5 @@ authors:
 - EEG channels couldn't be used as "virtual" EOG channels during ICA artifact
   detection. Reported by "fraenni" on the forum. Thank you! 🌻
   ({{ gh(572) }} by {{ authors.hoechenberger }})
+- Fix bug with handling of split files during preprocessing
+  ({{ gh(597) }} by {{ authors.larsoner }})
diff --git a/run.py b/run.py
index 421787830..c9bb2c503 100755
--- a/run.py
+++ b/run.py
@@ -15,6 +15,10 @@
 import coloredlogs
 
 
+# Ensure that the "scripts" that we import from is the correct one
+sys.path.insert(0, str(pathlib.Path(__file__).parent))
+
+
 logger = logging.getLogger(__name__)
 
 log_level_styles = {
diff --git a/scripts/preprocessing/_01_maxfilter.py b/scripts/preprocessing/_01_maxfilter.py
index efcb984e1..0c021b19a 100644
--- a/scripts/preprocessing/_01_maxfilter.py
+++ b/scripts/preprocessing/_01_maxfilter.py
@@ -28,7 +28,8 @@
 
 import config
 from config import (gen_log_kwargs, on_error, failsafe_run,
-                    import_experimental_data, import_er_data, import_rest_data)
+                    import_experimental_data, import_er_data, import_rest_data,
+                    _update_for_splits)
 from config import parallel_func
 
 logger = logging.getLogger('mne-bids-pipeline')
@@ -81,7 +82,6 @@ def run_maxwell_filter(*, cfg, subject, session=None, run=None, in_files=None):
         raise ValueError(f'You cannot set use_maxwell_filter to True '
                          f'if data have already processed with Maxwell-filter.'
                          f' Got proc={config.proc}.')
-
     bids_path_in = in_files[f"raw_run-{run}"]
     bids_path_out = bids_path_in.copy().update(
         processing="sss",
@@ -154,13 +154,16 @@ def run_maxwell_filter(*, cfg, subject, session=None, run=None, in_files=None):
     logger.info(**gen_log_kwargs(
         message=msg, subject=subject, session=session, run=run))
     raw_sss.save(out_files['sss_raw'], picks=picks, split_naming='bids',
-                 overwrite=True)
+                 overwrite=True, split_size=cfg._raw_split_size)
+    # we need to be careful about split files
+    _update_for_splits(out_files, 'sss_raw')
     del raw, raw_sss
 
     if cfg.interactive:
         # Load the data we have just written, because it contains only
         # the relevant channels.
-        raw_sss = mne.io.read_raw_fif(bids_path_out, allow_maxshield=True)
+        raw_sss = mne.io.read_raw_fif(
+            out_files['sss_raw'], allow_maxshield=True)
         raw_sss.plot(n_channels=50, butterfly=True, block=True)
         del raw_sss
 
@@ -211,7 +214,7 @@ def run_maxwell_filter(*, cfg, subject, session=None, run=None, in_files=None):
         # copy the bad channel selection from the reference run over to
         # the resting-state recording.
 
-        raw_sss = mne.io.read_raw_fif(bids_path_out)
+        raw_sss = mne.io.read_raw_fif(out_files['sss_raw'])
         rank_exp = mne.compute_rank(raw_sss, rank='info')['meg']
         rank_noise = mne.compute_rank(raw_noise_sss, rank='info')['meg']
 
@@ -248,8 +251,9 @@ def run_maxwell_filter(*, cfg, subject, session=None, run=None, in_files=None):
             message=msg, subject=subject, session=session, run=run))
         raw_noise_sss.save(
             out_files['sss_noise'], picks=picks, overwrite=True,
-            split_naming='bids'
+            split_naming='bids', split_size=cfg._raw_split_size,
         )
+        _update_for_splits(out_files, 'sss_noise')
         del raw_noise_sss
     return {key: pth.fpath for key, pth in out_files.items()}
 
@@ -290,6 +294,7 @@ def get_config(
         min_break_duration=config.min_break_duration,
         t_break_annot_start_after_previous_event=config.t_break_annot_start_after_previous_event,  # noqa:E501
         t_break_annot_stop_before_next_event=config.t_break_annot_stop_before_next_event,  # noqa:E501
+        _raw_split_size=config._raw_split_size,
     )
     return cfg
 
diff --git a/scripts/preprocessing/_02_frequency_filter.py b/scripts/preprocessing/_02_frequency_filter.py
index 3ab23ecea..1aae83de3 100644
--- a/scripts/preprocessing/_02_frequency_filter.py
+++ b/scripts/preprocessing/_02_frequency_filter.py
@@ -35,7 +35,8 @@
 
 import config
 from config import (gen_log_kwargs, on_error, failsafe_run,
-                    import_experimental_data, import_er_data, import_rest_data)
+                    import_experimental_data, import_er_data, import_rest_data,
+                    _update_for_splits)
 from config import parallel_func
 
 
@@ -75,11 +76,10 @@ def get_input_fnames_frequency_filter(**kwargs):
 
     if cfg.use_maxwell_filter:
         bids_path_in.update(processing="sss")
-        if bids_path_in.copy().update(split='01').fpath.exists():
-            bids_path_in = bids_path_in.update(split='01')
 
     in_files = dict()
     in_files[f'raw_run-{run}'] = bids_path_in
+    _update_for_splits(in_files, f'raw_run-{run}', single=True)
 
     if (cfg.process_er or config.noise_cov == 'rest') and run == cfg.runs[0]:
         noise_task = "rest" if config.noise_cov == "rest" else "noise"
@@ -87,9 +87,8 @@ def get_input_fnames_frequency_filter(**kwargs):
             raw_noise_fname_in = bids_path_in.copy().update(
                 run=None, task=noise_task
             )
-            if raw_noise_fname_in.copy().update(split='01').fpath.exists():
-                raw_noise_fname_in.update(split='01')
             in_files["raw_noise"] = raw_noise_fname_in
+            _update_for_splits(in_files, "raw_noise", single=True)
         else:
             if config.noise_cov == 'rest':
                 in_files["raw_rest"] = bids_path_in.copy().update(
@@ -189,7 +188,7 @@ def filter_data(
 
     out_files['raw_filt'] = bids_path.copy().update(
         root=cfg.deriv_root, processing='filt', extension='.fif',
-        suffix='raw')
+        suffix='raw', split=None)
     raw.load_data()
     filter(
         raw=raw, subject=subject, session=session, run=run,
@@ -201,7 +200,9 @@ def filter_data(
     resample(raw=raw, subject=subject, session=session, run=run,
              sfreq=cfg.resample_sfreq, data_type='experimental')
 
-    raw.save(out_files['raw_filt'], overwrite=True, split_naming='bids')
+    raw.save(out_files['raw_filt'], overwrite=True, split_naming='bids',
+             split_size=cfg._raw_split_size)
+    _update_for_splits(out_files, 'raw_filt')
     if cfg.interactive:
         # Plot raw data and power spectral density.
         raw.plot(n_channels=50, butterfly=True)
@@ -238,7 +239,7 @@ def filter_data(
         out_files['raw_noise_filt'] = \
             bids_path_noise.copy().update(
                 root=cfg.deriv_root, processing='filt', extension='.fif',
-                suffix='raw')
+                suffix='raw', split=None)
 
         raw_noise.load_data()
         filter(
@@ -252,8 +253,10 @@ def filter_data(
                  sfreq=cfg.resample_sfreq, data_type=data_type)
 
         raw_noise.save(
-            out_files['raw_noise_filt'], overwrite=True, split_naming='bids'
+            out_files['raw_noise_filt'], overwrite=True, split_naming='bids',
+            split_size=cfg._raw_split_size,
         )
+        _update_for_splits(out_files, 'raw_noise_filt')
         if cfg.interactive:
             # Plot raw data and power spectral density.
             raw_noise.plot(n_channels=50, butterfly=True)
@@ -301,6 +304,7 @@ def get_config(
         min_break_duration=config.min_break_duration,
         t_break_annot_start_after_previous_event=config.t_break_annot_start_after_previous_event,  # noqa:E501
         t_break_annot_stop_before_next_event=config.t_break_annot_stop_before_next_event,  # noqa:E501
+        _raw_split_size=config._raw_split_size,
     )
     return cfg
 
diff --git a/scripts/preprocessing/_03_make_epochs.py b/scripts/preprocessing/_03_make_epochs.py
index 7a70b45db..6e5c4198c 100644
--- a/scripts/preprocessing/_03_make_epochs.py
+++ b/scripts/preprocessing/_03_make_epochs.py
@@ -20,7 +20,7 @@
 
 import config
 from config import make_epochs, gen_log_kwargs, on_error, failsafe_run
-from config import parallel_func
+from config import parallel_func, _update_for_splits
 
 logger = logging.getLogger('mne-bids-pipeline')
 
@@ -44,10 +44,7 @@ def run_epochs(*, cfg, subject, session=None):
     for run in cfg.runs:
         raw_fname_in = bids_path.copy().update(run=run, processing='filt',
                                                suffix='raw', check=False)
-
-        if raw_fname_in.copy().update(split='01').fpath.exists():
-            raw_fname_in.update(split='01')
-
+        raw_fname_in = _update_for_splits(raw_fname_in, None, single=True)
         raw_fnames.append(raw_fname_in)
 
     # Generate a unique event name -> event code mapping that can be used
@@ -191,7 +188,10 @@ def run_epochs(*, cfg, subject, session=None):
     logger.info(**gen_log_kwargs(message=msg, subject=subject,
                                  session=session))
     epochs_fname = bids_path.copy().update(suffix='epo', check=False)
-    epochs.save(epochs_fname, overwrite=True, split_naming='bids')
+    epochs.save(
+        epochs_fname, overwrite=True, split_naming='bids',
+        split_size=cfg._epochs_split_size)
+    # _update_for_splits(out_files, 'epochs')
 
     if cfg.interactive:
         epochs.plot()
@@ -228,7 +228,8 @@ def get_config(
         event_repeated=config.event_repeated,
         decim=config.decim,
         ch_types=config.ch_types,
-        eeg_reference=config.get_eeg_reference()
+        eeg_reference=config.get_eeg_reference(),
+        _epochs_split_size=config._epochs_split_size,
     )
     return cfg
 
diff --git a/scripts/preprocessing/_04a_run_ica.py b/scripts/preprocessing/_04a_run_ica.py
index 308371c9a..aa5967407 100644
--- a/scripts/preprocessing/_04a_run_ica.py
+++ b/scripts/preprocessing/_04a_run_ica.py
@@ -33,7 +33,7 @@
 
 import config
 from config import (make_epochs, gen_log_kwargs, on_error, failsafe_run,
-                    annotations_to_events)
+                    annotations_to_events, _update_for_splits)
 from config import parallel_func
 
 
@@ -259,9 +259,7 @@ def run_ica(*, cfg, subject, session=None):
     raw_fnames = []
     for run in cfg.runs:
         raw_fname.update(run=run)
-        if raw_fname.copy().update(split='01').fpath.exists():
-            raw_fname.update(split='01')
-
+        raw_fname = _update_for_splits(raw_fname, None, single=True)
         raw_fnames.append(raw_fname.copy())
 
     # Generate a unique event name -> event code mapping that can be used
diff --git a/scripts/preprocessing/_04b_run_ssp.py b/scripts/preprocessing/_04b_run_ssp.py
index 999b121fb..cd1e6a8c1 100644
--- a/scripts/preprocessing/_04b_run_ssp.py
+++ b/scripts/preprocessing/_04b_run_ssp.py
@@ -18,7 +18,7 @@
 
 import config
 from config import gen_log_kwargs, on_error, failsafe_run
-from config import parallel_func
+from config import parallel_func, _update_for_splits
 
 
 logger = logging.getLogger('mne-bids-pipeline')
@@ -49,9 +49,7 @@ def run_ssp(*, cfg, subject, session=None):
     msg = f'Input: {raw_fname_in.basename}, Output: {proj_fname_out.basename}'
     logger.info(**gen_log_kwargs(message=msg, subject=subject,
                                  session=session))
-
-    if raw_fname_in.copy().update(split='01').fpath.exists():
-        raw_fname_in.update(split='01')
+    raw_fname_in = _update_for_splits(raw_fname_in, None, single=True)
 
     raw = mne.io.read_raw_fif(raw_fname_in)
     msg = 'Computing SSPs for ECG'
diff --git a/scripts/preprocessing/_05a_apply_ica.py b/scripts/preprocessing/_05a_apply_ica.py
index a0ce08af4..170b15fc4 100644
--- a/scripts/preprocessing/_05a_apply_ica.py
+++ b/scripts/preprocessing/_05a_apply_ica.py
@@ -93,7 +93,10 @@ def apply_ica(*, cfg, subject, session):
     msg = 'Saving reconstructed epochs after ICA.'
     logger.info(**gen_log_kwargs(message=msg, subject=subject,
                                  session=session))
-    epochs_cleaned.save(fname_epo_out, overwrite=True, split_naming='bids')
+    epochs_cleaned.save(
+        fname_epo_out, overwrite=True, split_naming='bids',
+        split_size=cfg._epochs_split_size)
+    # _update_for_splits(out_files, 'epochs_cleaned')
 
     # Compare ERP/ERF before and after ICA artifact rejection. The evoked
     # response is calculated across ALL epochs, just like ICA was run on
@@ -127,7 +130,8 @@ def get_config(
         deriv_root=config.get_deriv_root(),
         interactive=config.interactive,
         baseline=config.baseline,
-        ica_reject=config.get_ica_reject()
+        ica_reject=config.get_ica_reject(),
+        _epochs_split_size=config._epochs_split_size,
     )
     return cfg
 
diff --git a/scripts/preprocessing/_05b_apply_ssp.py b/scripts/preprocessing/_05b_apply_ssp.py
index 9c693cba4..c13c1ad07 100644
--- a/scripts/preprocessing/_05b_apply_ssp.py
+++ b/scripts/preprocessing/_05b_apply_ssp.py
@@ -62,7 +62,10 @@ def apply_ssp(*, cfg, subject, session=None):
     msg = 'Saving epochs with projectors.'
     logger.info(**gen_log_kwargs(message=msg, subject=subject,
                                  session=session))
-    epochs_cleaned.save(fname_out, overwrite=True, split_naming='bids')
+    epochs_cleaned.save(
+        fname_out, overwrite=True, split_naming='bids',
+        split_size=cfg._epochs_split_size)
+    # _update_for_splits(out_files, 'epochs_cleaned')
 
 
 def get_config(
@@ -76,6 +79,7 @@ def get_config(
         rec=config.rec,
         space=config.space,
         deriv_root=config.get_deriv_root(),
+        _epochs_split_size=config._epochs_split_size,
     )
     return cfg
 
diff --git a/scripts/preprocessing/_06_ptp_reject.py b/scripts/preprocessing/_06_ptp_reject.py
index 5c8c4537c..8d8439730 100644
--- a/scripts/preprocessing/_06_ptp_reject.py
+++ b/scripts/preprocessing/_06_ptp_reject.py
@@ -89,7 +89,9 @@ def drop_ptp(*, cfg, subject, session=None):
     msg = 'Saving cleaned, baseline-corrected epochs …'
 
     epochs.apply_baseline(cfg.baseline)
-    epochs.save(fname_out, overwrite=True, split_naming='bids')
+    epochs.save(
+        fname_out, overwrite=True, split_naming='bids',
+        split_size=cfg._epochs_split_size)
 
 
 def get_config(
@@ -108,7 +110,8 @@ def get_config(
         spatial_filter=config.spatial_filter,
         ica_reject=config.get_ica_reject(),
         deriv_root=config.get_deriv_root(),
-        decim=config.decim
+        decim=config.decim,
+        _epochs_split_size=config._epochs_split_size,
     )
     return cfg
 
diff --git a/scripts/report/_01_make_reports.py b/scripts/report/_01_make_reports.py
index 67bd5cfd4..3fee6e83c 100644
--- a/scripts/report/_01_make_reports.py
+++ b/scripts/report/_01_make_reports.py
@@ -26,7 +26,7 @@
 import config
 from config import (
     gen_log_kwargs, on_error, failsafe_run, parallel_func,
-    get_noise_cov_bids_path
+    get_noise_cov_bids_path, _update_for_splits,
 )
 
 
@@ -52,10 +52,7 @@ def get_events(cfg, subject, session):
 
     for run in cfg.runs:
         this_raw_fname = raw_fname.copy().update(run=run)
-
-        if this_raw_fname.copy().update(split='01').fpath.exists():
-            this_raw_fname.update(split='01')
-
+        this_raw_fname = _update_for_splits(this_raw_fname, None, single=True)
         raw_filt = mne.io.read_raw_fif(this_raw_fname)
         raws_filt.append(raw_filt)
         del this_raw_fname
@@ -81,10 +78,7 @@ def get_er_path(cfg, subject, session):
                          datatype=cfg.datatype,
                          root=cfg.deriv_root,
                          check=False)
-
-    if raw_fname.copy().update(split='01').fpath.exists():
-        raw_fname.update(split='01')
-
+    raw_fname = _update_for_splits(raw_fname, None, single=True)
     return raw_fname
 
 
@@ -482,9 +476,7 @@ def run_report_preprocessing(
             run=run, processing='filt',
             suffix='raw', check=False
         )
-        if fname.copy().update(split='01').fpath.exists():
-            fname.update(split='01')
-
+        fname = _update_for_splits(fname, None, single=True)
         fnames_raw_filt.append(fname)
 
     fname_epo_not_clean = bids_path.copy().update(suffix='epo')
diff --git a/tests/configs/config_ds000248.py b/tests/configs/config_ds000248.py
index c4b2e9401..098aef3d5 100644
--- a/tests/configs/config_ds000248.py
+++ b/tests/configs/config_ds000248.py
@@ -24,6 +24,8 @@
 find_noisy_channels_meg = True
 use_maxwell_filter = True
 process_er = True
+_raw_split_size = '60MB'  # hits both task-noise and task-audiovisual
+_epochs_split_size = '30MB'
 
 
 def noise_cov(bp):
diff --git a/tests/configs/config_ds003775.py b/tests/configs/config_ds003775.py
index ed5ffa493..b317cc40d 100644
--- a/tests/configs/config_ds003775.py
+++ b/tests/configs/config_ds003775.py
@@ -37,7 +37,6 @@
 parallel_backend = 'loky'
 dask_open_dashboard = True
 
-on_error = 'continue'
 log_level = 'info'
 
 N_JOBS = 1