diff --git a/CHANGELOG.md b/CHANGELOG.md
index aba3b3551..c9c068667 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,8 @@
 
 ## Unreleased
 
+- Using pytest-xdist for faster local tests
+  [PR #440](https://github.com/aai-institute/pyDVL/pull/440)
 - Added `AntitheticPermutationSampler`
   [PR #439](https://github.com/aai-institute/pyDVL/pull/439)
 - Implementation of Data-OOB by @BastienZim
diff --git a/requirements-dev.txt b/requirements-dev.txt
index deeabebee..6bf514bb2 100644
--- a/requirements-dev.txt
+++ b/requirements-dev.txt
@@ -19,5 +19,6 @@ pytest-docker==2.0.0
 pytest-mock
 pytest-timeout
 pytest-lazy-fixture
+pytest-xdist>=3.3.1
 wheel
 twine==4.0.2
diff --git a/tests/conftest.py b/tests/conftest.py
index c99a38d3f..19ad9bf7a 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -196,7 +196,6 @@ def seed_numpy(seed=42):
     np.random.seed(seed)
 
 
-@pytest.fixture(scope="session")
 def num_workers():
     # Run with 2 CPUs inside GitHub actions
     if os.getenv("CI"):
@@ -205,9 +204,22 @@ def num_workers():
     return max(1, min(available_cpus() - 1, 4))
 
 
-@pytest.fixture
-def n_jobs(num_workers):
-    return num_workers
+@pytest.fixture(scope="session")
+def n_jobs():
+    return num_workers()
+
+
+def pytest_xdist_auto_num_workers(config) -> Optional[int]:
+    """Return the number of workers to use for pytest-xdist.
+
+    This is used by pytest-xdist to automatically determine the number of
+    workers to use. We want to use all available CPUs, but leave one CPU for
+    the main process.
+    """
+
+    if config.option.numprocesses == "auto":
+        return max(1, (available_cpus() - 1) // num_workers())
+    return None
 
 
 ################################################################################
diff --git a/tests/utils/conftest.py b/tests/utils/conftest.py
index 19c8a0798..276972c73 100644
--- a/tests/utils/conftest.py
+++ b/tests/utils/conftest.py
@@ -2,17 +2,19 @@
 
 from pydvl.parallel.config import ParallelConfig
 
+from ..conftest import num_workers
+
 
 @pytest.fixture(scope="module", params=["joblib", "ray-local", "ray-external"])
-def parallel_config(request, num_workers):
+def parallel_config(request):
     if request.param == "joblib":
-        yield ParallelConfig(backend="joblib", n_cpus_local=num_workers)
+        yield ParallelConfig(backend="joblib", n_cpus_local=num_workers())
     elif request.param == "ray-local":
         try:
             import ray
         except ImportError:
             pytest.skip("Ray not installed.")
-        yield ParallelConfig(backend="ray", n_cpus_local=num_workers)
+        yield ParallelConfig(backend="ray", n_cpus_local=num_workers())
         ray.shutdown()
     elif request.param == "ray-external":
         try:
@@ -22,10 +24,7 @@ def parallel_config(request, num_workers):
             pytest.skip("Ray not installed.")
         # Starts a head-node for the cluster.
         cluster = Cluster(
-            initialize_head=True,
-            head_node_args={
-                "num_cpus": num_workers,
-            },
+            initialize_head=True, head_node_args={"num_cpus": num_workers()}
         )
         yield ParallelConfig(backend="ray", address=cluster.address)
         ray.shutdown()
diff --git a/tests/utils/test_parallel.py b/tests/utils/test_parallel.py
index 4677de4ee..bca0c56af 100644
--- a/tests/utils/test_parallel.py
+++ b/tests/utils/test_parallel.py
@@ -12,15 +12,17 @@
 from pydvl.parallel.futures import init_executor
 from pydvl.utils.types import Seed
 
+from ..conftest import num_workers
 
-def test_effective_n_jobs(parallel_config, num_workers):
+
+def test_effective_n_jobs(parallel_config):
     parallel_backend = init_parallel_backend(parallel_config)
     assert parallel_backend.effective_n_jobs(1) == 1
-    assert parallel_backend.effective_n_jobs(4) == min(4, num_workers)
+    assert parallel_backend.effective_n_jobs(4) == min(4, num_workers())
     if parallel_config.address is None:
-        assert parallel_backend.effective_n_jobs(-1) == num_workers
+        assert parallel_backend.effective_n_jobs(-1) == num_workers()
     else:
-        assert parallel_backend.effective_n_jobs(-1) == num_workers
+        assert parallel_backend.effective_n_jobs(-1) == num_workers()
 
     for n_jobs in [-1, 1, 2]:
         assert parallel_backend.effective_n_jobs(n_jobs) == effective_n_jobs(
@@ -166,7 +168,7 @@ def test_map_reduce_seeding(parallel_config, seed_1, seed_2, op):
     assert op(result_1, result_2)
 
 
-def test_wrap_function(parallel_config, num_workers):
+def test_wrap_function(parallel_config):
     if parallel_config.backend != "ray":
         pytest.skip("Only makes sense for ray")
 
@@ -188,8 +190,8 @@ def get_pid():
         return os.getpid()
 
     wrapped_func = parallel_backend.wrap(get_pid, num_cpus=1)
-    pids = parallel_backend.get([wrapped_func() for _ in range(num_workers)])
-    assert len(set(pids)) == num_workers
+    pids = parallel_backend.get([wrapped_func() for _ in range(num_workers())])
+    assert len(set(pids)) == num_workers()
 
 
 def test_futures_executor_submit(parallel_config):
@@ -205,7 +207,7 @@ def test_futures_executor_map(parallel_config):
     assert results == [1, 2, 3]
 
 
-def test_futures_executor_map_with_max_workers(parallel_config, num_workers):
+def test_futures_executor_map_with_max_workers(parallel_config):
     if parallel_config.backend != "ray":
         pytest.skip("Currently this test only works with Ray")
 
@@ -215,12 +217,12 @@ def func(_):
 
     start_time = time.monotonic()
     with init_executor(config=parallel_config) as executor:
-        assert executor._max_workers == num_workers
+        assert executor._max_workers == num_workers()
         list(executor.map(func, range(3)))
     end_time = time.monotonic()
     total_time = end_time - start_time
-    # We expect the time difference to be > 3 / num_workers, but has to be at least 1
-    assert total_time > max(1.0, 3 / num_workers)
+    # We expect the time difference to be > 3 / num_workers(), but has to be at least 1
+    assert total_time > max(1.0, 3 / num_workers())
 
 
 def test_future_cancellation(parallel_config):
diff --git a/tests/value/conftest.py b/tests/value/conftest.py
index 38cddfdf8..75fedb38d 100644
--- a/tests/value/conftest.py
+++ b/tests/value/conftest.py
@@ -10,6 +10,7 @@
 from pydvl.utils.status import Status
 from pydvl.value import ValuationResult
 
+from ..conftest import num_workers
 from . import polynomial
 
 
@@ -122,5 +123,5 @@ def linear_shapley(linear_dataset, scorer, n_jobs):
 
 
 @pytest.fixture(scope="module")
-def parallel_config(num_workers):
-    yield ParallelConfig(backend="joblib", n_cpus_local=num_workers, wait_timeout=0.1)
+def parallel_config():
+    yield ParallelConfig(backend="joblib", n_cpus_local=num_workers(), wait_timeout=0.1)
diff --git a/tox.ini b/tox.ini
index b01f1dfb5..6b6e14c75 100644
--- a/tox.ini
+++ b/tox.ini
@@ -12,12 +12,12 @@ setenv =
 [testenv:base]
 description = Tests base modules
 commands =
-    pytest --cov "{envsitepackagesdir}/pydvl" -m "not torch" {posargs}
+    pytest -n auto --cov "{envsitepackagesdir}/pydvl" -m "not torch" {posargs}
 
 [testenv:torch]
 description = Tests modules that rely on pytorch
 commands =
-    pytest --cov "{envsitepackagesdir}/pydvl" -m torch {posargs}
+    pytest -n auto --cov "{envsitepackagesdir}/pydvl" -m torch {posargs}
 extras =
     influence
 
@@ -26,7 +26,7 @@ description = Tests notebooks
 setenv =
     PYTHONPATH={toxinidir}/notebooks
 commands =
-    pytest notebooks/ --cov "{envsitepackagesdir}/pydvl"
+    pytest -n auto notebooks/ --cov "{envsitepackagesdir}/pydvl" {posargs}
 deps =
     {[testenv]deps}
     jupyter==1.0.0