diff --git a/tests/conftest.py b/tests/conftest.py
new file mode 100644
index 0000000..6e3e3f2
--- /dev/null
+++ b/tests/conftest.py
@@ -0,0 +1,75 @@
+import pytest
+import numpy as np
+import scipy.sparse as sp
+
+from lightfm import LightFM
+
+# set our default random seed
+SEED = 42
+
+
+@pytest.fixture(scope="session")
+def rng():
+    """Initialise a shared random number generator for all tests."""
+
+    return np.random.RandomState(SEED)
+
+
+@pytest.fixture(scope="session")
+def array_int32(rng, size=10):
+    """Initialise an array of type np.int32 of size `size`."""
+
+    return rng.randint(0, 100, size=size, dtype=np.int32)
+
+
+@pytest.fixture(
+    scope="session", ids=["tuple", "list", "ndarray"], params=[tuple, list, np.array]
+)
+def user_ids(array_int32, request):
+    """Initialise input user_ids valid for calls to the LightFM.predict method.
+
+    Notes
+    -----
+    On parameterized pytest fixtures: This fixture will iterate over all passed
+    `params`. This avoids having to apply a `pytest.mark.parameterize` decorator to
+    every test that needs the same `user_ids`.
+
+    You can find out more about parameterized fixtures in the pytest docs:
+    https://docs.pytest.org/en/stable/parametrize.html
+
+    """
+
+    _type = request.param
+    yield _type(array_int32)
+
+
+@pytest.fixture(
+    scope="session", ids=["tuple", "list", "ndarray"], params=[tuple, list, np.array]
+)
+def item_ids(array_int32, request):
+    """Initialise input item_ids valid for calls to the LightFM.predict method.
+
+    Notes
+    -----
+    See `user_ids` fixture for a note on parameterized fixtures.
+
+    """
+    _type = request.param
+    yield _type(array_int32)
+
+
+@pytest.fixture(scope="session")
+def train_matrix(rng, n_users=1000, n_items=1000):
+    """Create a random sparse CSR matrix of shape (n_users, n_items) for training."""
+
+    return sp.rand(n_users, n_items, format="csr", random_state=rng)
+
+
+@pytest.fixture(scope="session")
+def lfm(train_matrix):
+    """Create a _trained_ LightFM model instance."""
+
+    model = LightFM()
+    model.fit(train_matrix)
+
+    return model
diff --git a/tests/test_api.py b/tests/test_api.py
index 2863212..54ad46f 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -380,3 +380,16 @@ def test_warp_few_items():
     model = LightFM(loss="warp", max_sampled=10)
 
     model.fit(train)
+
+
+def test_predict_user_item_inputs_with_valid_types(lfm, user_ids, item_ids):
+    """Test that calls to the predict method with inputs of valid types succeed."""
+
+    # GIVEN user_ids of a valid type
+    # AND item_ids of a valid type
+    # WHEN trained model provided
+    # THEN calls to LightFM.predict succeed
+
+    h = lfm.predict(user_ids=user_ids, item_ids=item_ids)
+    assert h.dtype == np.float32
+    assert len(h) == len(user_ids)