diff --git a/src/pyrecest/filters/_ukf.py b/src/pyrecest/filters/_ukf.py
index f09c393b9..b5225367c 100644
--- a/src/pyrecest/filters/_ukf.py
+++ b/src/pyrecest/filters/_ukf.py
@@ -116,11 +116,19 @@ def predict(self, fx=None, dt=None, **fx_args):
 
         x_pred = einsum("i,ij->j", Wm, sigmas_f)
 
-        P_pred = zeros((self._model.dim_x, self._model.dim_x))
+        process_covariance = asarray(self.Q, dtype=float64)
+        expected_process_shape = (self._model.dim_x, self._model.dim_x)
+        if process_covariance.shape != expected_process_shape:
+            raise ValueError(
+                "process noise covariance Q has shape "
+                f"{process_covariance.shape}, expected {expected_process_shape}"
+            )
+
+        P_pred = zeros(expected_process_shape)
         for i in range(n_sigmas):
             d = expand_dims(sigmas_f[i] - x_pred, -1)
             P_pred = P_pred + Wc[i] * (d @ transpose(d))
-        P_pred = P_pred + asarray(self.Q, dtype=float64)
+        P_pred = P_pred + process_covariance
         P_pred = 0.5 * (P_pred + transpose(P_pred))
 
         self.x = x_pred
diff --git a/tests/filters/test_ukf_process_noise_shape.py b/tests/filters/test_ukf_process_noise_shape.py
new file mode 100644
index 000000000..4ec7273c1
--- /dev/null
+++ b/tests/filters/test_ukf_process_noise_shape.py
@@ -0,0 +1,35 @@
+import unittest
+
+import numpy.testing as npt
+
+# pylint: disable=no-name-in-module,no-member
+import pyrecest.backend
+from pyrecest.backend import array, diag
+from pyrecest.distributions import GaussianDistribution
+from pyrecest.filters.unscented_kalman_filter import UnscentedKalmanFilter
+
+
+class UnscentedKalmanFilterProcessNoiseShapeTest(unittest.TestCase):
+    @unittest.skipIf(
+        pyrecest.backend.__backend_name__ in ("pytorch", "jax"),
+        reason="Not supported on this backend",
+    )
+    def test_vector_process_covariance_is_rejected_without_state_mutation(self):
+        initial_mean = array([0.5, -0.25])
+        initial_covariance = diag(array([1.2, 0.8]))
+        ukf = UnscentedKalmanFilter(
+            GaussianDistribution(initial_mean, initial_covariance)
+        )
+
+        with self.assertRaisesRegex(
+            ValueError,
+            r"process noise covariance Q has shape .* expected \(2, 2\)",
+        ):
+            ukf.predict_identity(array([0.4, 0.2]))
+
+        npt.assert_allclose(ukf.get_point_estimate(), initial_mean)
+        npt.assert_allclose(ukf.filter_state.covariance(), initial_covariance)
+
+
+if __name__ == "__main__":
+    unittest.main()