diff --git a/tests/codec_be_on_mr_selection/__init__.py b/tests/codec_be_on_mr_selection/__init__.py
index 34a307a7393396bbb6aa614f4d496e25c2757bc2..b7f4cd8efeffaff30d451bb9d0eb49945f68a47d 100644
--- a/tests/codec_be_on_mr_selection/__init__.py
+++ b/tests/codec_be_on_mr_selection/__init__.py
@@ -35,6 +35,7 @@ import filecmp
 from pathlib import Path
 import subprocess
 from .constants import OUTPUT_MODES_AND_OPTIONS_FOR_EXPERIMENT, DTX_ON, FER_5PERC
+from ..testconfig import MD5_REF_DICT
 
 HERE = Path(__file__).parent
 # set environment variables in CI job
@@ -98,8 +99,20 @@ def apply_error_pattern_on_bitstream(
         subprocess.run(cmd)
 
 
-def files_equal(dut_file, ref_file):
-    return filecmp.cmp(dut_file, ref_file)
+def is_be_to_ref(dut_file):
+    """
+    Check bitexactness either by comparing files directly or by comparing MD5 sums
+    """
+    if MD5_REF_DICT == dict():
+        ref_file = REF_PATH.joinpath(dut_file.name)
+        is_be = filecmp.cmp(dut_file, ref_file)
+    else:
+        md5_ref = MD5_REF_DICT[dut_file.name]
+        cmd = f"powershell.exe (Get-FileHash {str(dut_file)} -Algorithm MD5).Hash"
+        md5_dut = subprocess.check_output(cmd, shell=True).decode().splitlines()[-1]
+        is_be = md5_ref == md5_dut
+
+    return is_be
 
 
 def run_check(
@@ -113,6 +126,7 @@ def run_check(
     decoder_frontend,
     is_ref_creation,
     input_file_num=None,
+    keep_files=True,
 ):
     sampling_rate = 48
     output_mode, options = OUTPUT_MODES_AND_OPTIONS_FOR_EXPERIMENT[experiment]
@@ -147,9 +161,8 @@ def run_check(
         add_option_list=options + [str(f) for f in metadata],
     )
 
-    ref_bitstream = REF_PATH.joinpath(dut_bitstream.name)
-    if not is_ref_creation and not files_equal(dut_bitstream, ref_bitstream):
-        pytest.fail("Bitstream file differs from reference")
+    if not is_be_to_ref(dut_bitstream):
+        pytest.fail(f"Bitstream file differs from reference")
 
     dut_bitstream_to_decoder = dut_bitstream
     if error_pattern is not None:
@@ -168,12 +181,15 @@ def run_check(
     #       this should not be a problem as both the reference and the tdut output was generated by the codec, so
     #       diverging headers should also indicate a problem - still, keep in mind if something bogus happens
     if not is_ref_creation:
-        ref_output = REF_PATH.joinpath(dut_output.name)
-        if not files_equal(dut_output, ref_output):
+        if not is_be_to_ref(dut_output):
             pytest.fail("Decoder output differs from reference")
+        elif not keep_files:
+            os.remove(dut_output)
+            os.remove(dut_bitstream)
         for md in metadata:
             md_suffix = "".join(md.suffixes)
             dut_md = DUT_PATH.joinpath(dut_output.with_suffix(md_suffix).name)
-            ref_md = REF_PATH.joinpath(dut_output.with_suffix(md_suffix).name)
-            if not files_equal(dut_md, ref_md):
+            if not is_be_to_ref(dut_md):
                 pytest.fail("Metadata file {md.name} differs from reference")
+            elif not keep_files:
+                os.remove(dut_md)
diff --git a/tests/codec_be_on_mr_selection/test_experiments.py b/tests/codec_be_on_mr_selection/test_experiments.py
index a0a4aa780a757cf887e6b80e72a8a48eed00899d..5b2f51689342117c2e543442a974bce6714ff6fc 100644
--- a/tests/codec_be_on_mr_selection/test_experiments.py
+++ b/tests/codec_be_on_mr_selection/test_experiments.py
@@ -52,6 +52,7 @@ def test_p800(
     dut_encoder_frontend,
     dut_decoder_frontend,
     update_ref,
+    keep_files,
 ):
     run_check(
         experiment,
@@ -63,6 +64,7 @@ def test_p800(
         dut_encoder_frontend,
         dut_decoder_frontend,
         update_ref == 1,
+        keep_files=keep_files,
     )
 
 
@@ -79,6 +81,7 @@ def test_bs1534_no_masa(
     dut_encoder_frontend,
     dut_decoder_frontend,
     update_ref,
+    keep_files,
 ):
     category = ""
     run_check(
@@ -92,6 +95,7 @@ def test_bs1534_no_masa(
         dut_decoder_frontend,
         update_ref == 1,
         input_file_num=input_file_num,
+        keep_files=keep_files,
     )
 
 
@@ -111,6 +115,7 @@ def test_bs1534_masa(
     dut_encoder_frontend,
     dut_decoder_frontend,
     update_ref,
+    keep_files,
 ):
     run_check(
         experiment,
@@ -123,4 +128,5 @@ def test_bs1534_masa(
         dut_decoder_frontend,
         update_ref == 1,
         input_file_num=input_file_num,
+        keep_files=keep_files,
     )
diff --git a/tests/conftest.py b/tests/conftest.py
index 12b62caccd9d49cd003b204cfe8ce7e307674a24..e4de453b0ea6698f9600076ded65a53a1a2b7d51 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -133,6 +133,12 @@ def pytest_addoption(parser):
         " Use --keep_files to prevent these deletions.",
     )
 
+    parser.addoption(
+        "--selection_be_md5_file",
+        type=Path,
+        help="Path to file with md5 sums for the reference signals of the selection-BE test"
+    )
+
 
 @pytest.fixture(scope="session", autouse=True)
 def update_ref(request):
@@ -513,3 +519,9 @@ def pytest_configure(config):
     )
     if config.option.param_file:
         testconfig.PARAM_FILE = config.option.param_file
+    if config.option.selection_be_md5_file:
+        md5_file_path = config.option.selection_be_md5_file
+        if not platform.system() == "Windows":
+            raise NotImplementedError("MD5 comparison is currently hardcoded for windows")
+        with open(md5_file_path) as f:
+            testconfig.MD5_REF_DICT = {line.split()[0]: line.split()[1] for line in f.readlines()}
diff --git a/tests/testconfig.py b/tests/testconfig.py
index 12d170ba8db49171e16add1f827a590ffcb09bd3..1dbfbb840343801460f1548b4e8b662d9ce96e6e 100644
--- a/tests/testconfig.py
+++ b/tests/testconfig.py
@@ -35,3 +35,5 @@ To configure test modules.
 """
 
 PARAM_FILE = "scripts/config/self_test.prm"
+
+MD5_REF_DICT = dict()
\ No newline at end of file