diff --git a/qadence/backends/pyqtorch/backend.py b/qadence/backends/pyqtorch/backend.py
index 6378d365..e303533f 100644
--- a/qadence/backends/pyqtorch/backend.py
+++ b/qadence/backends/pyqtorch/backend.py
@@ -108,7 +108,7 @@ def circuit(self, circuit: QuantumCircuit) -> ConvertedCircuit:
         original_circ = circuit
         if len(passes) > 0:
             circuit = transpile(*passes)(circuit)
-        # setting noise on blocks
+        # Setting noise in the circuit.
         if self.config.noise:
             set_noise(circuit, self.config.noise)
 
@@ -136,8 +136,6 @@ def observable(self, observable: AbstractBlock, n_qubits: int) -> ConvertedObser
             scale_primitive_blocks_only,
         ]
         block = transpile(*transpilations)(observable)  # type: ignore[call-overload]
-        # we do not set noise on the observable blocks
-        # as this would not be correct when computing expectations
         operations = convert_block(block, n_qubits, self.config)
         native = pyq.Observable(operations=operations)
         return ConvertedObservable(native=native, abstract=block, original=observable)
diff --git a/tests/qadence/test_noise/test_digital_noise.py b/tests/qadence/test_noise/test_digital_noise.py
index 60f50363..d65409c7 100644
--- a/tests/qadence/test_noise/test_digital_noise.py
+++ b/tests/qadence/test_noise/test_digital_noise.py
@@ -104,20 +104,28 @@ def test_expectation_digital_noise(noisy_config: NoiseProtocol | list[NoiseProto
     circuit = QuantumCircuit(2, block)
     observable = hamiltonian_factory(circuit.n_qubits, detuning=Z)
     noise = NoiseHandler(noisy_config, {"error_probability": 0.1})
+    backend = backend_factory(backend=BackendName.PYQTORCH, diff_mode=DiffMode.AD)
 
     # Construct a quantum model.
     model = QuantumModel(circuit=circuit, observable=observable)
     noiseless_expectation = model.expectation(values={})
 
+    (pyqtorch_circ, pyqtorch_obs, embed, params) = backend.convert(circuit, observable)
+    native_noisy_expectation = backend.expectation(
+        pyqtorch_circ, pyqtorch_obs, embed(params, {}), noise=noise
+    )
+    assert not torch.allclose(noiseless_expectation, native_noisy_expectation)
+
     noisy_model = QuantumModel(circuit=circuit, observable=observable, noise=noise)
-    noisy_expectation = noisy_model.expectation(values={})
-    assert not torch.allclose(noiseless_expectation, noisy_expectation)
+    noisy_model_expectation = noisy_model.expectation(values={})
+    assert torch.allclose(noisy_model_expectation, native_noisy_expectation)
 
-    backend = backend_factory(backend=BackendName.PYQTORCH, diff_mode=DiffMode.AD)
     (pyqtorch_circ, pyqtorch_obs, embed, params) = backend.convert(circuit, observable)
-    native_expectation = backend.expectation(pyqtorch_circ, pyqtorch_obs, embed(params, {}))
+    noisy_converted_model_expectation = backend.expectation(
+        pyqtorch_circ, pyqtorch_obs, embed(params, {})
+    )
 
-    assert torch.allclose(noisy_expectation, native_expectation)
+    assert torch.allclose(noisy_converted_model_expectation, native_noisy_expectation)
 
 
 @pytest.mark.parametrize(