diff --git a/pyro/nn/__init__.py b/pyro/nn/__init__.py
index e55e7356f6..310805a7c2 100644
--- a/pyro/nn/__init__.py
+++ b/pyro/nn/__init__.py
@@ -14,6 +14,7 @@
     PyroModuleList,
     PyroParam,
     PyroSample,
+    PyroSamplePlateScope,
     pyro_method,
 )
 
@@ -28,4 +29,5 @@
     "PyroSample",
     "pyro_method",
     "PyroModuleList",
+    "PyroSamplePlateScope",
 ]
diff --git a/pyro/nn/module.py b/pyro/nn/module.py
index b84a17875b..d3eff1d823 100644
--- a/pyro/nn/module.py
+++ b/pyro/nn/module.py
@@ -37,6 +37,7 @@ def _copy_to_script_wrapper(fn):
     Any,
     Callable,
     Dict,
+    Iterable,
     Iterator,
     List,
     NamedTuple,
@@ -53,8 +54,10 @@ def _copy_to_script_wrapper(fn):
 
 import pyro
 import pyro.params.param_store
+from pyro.distributions.torch_distribution import TorchDistributionMixin
 from pyro.ops.provenance import detach_provenance
-from pyro.poutine.runtime import _PYRO_PARAM_STORE
+from pyro.poutine.messenger import Messenger
+from pyro.poutine.runtime import _PYRO_PARAM_STORE, InferDict
 
 _MODULE_LOCAL_PARAMS: bool = False
 
@@ -63,7 +66,6 @@ def _copy_to_script_wrapper(fn):
 _PyroModule = TypeVar("_PyroModule", bound="PyroModule")
 
 if TYPE_CHECKING:
-    from pyro.distributions.torch_distribution import TorchDistributionMixin
     from pyro.params.param_store import StateDict
 
 
@@ -232,6 +234,48 @@ def __get__(
         return value
 
 
+class _PyroSampleInferDict(InferDict):
+    _original_pyrosample_dist: TorchDistributionMixin
+
+
+class PyroSamplePlateScope(Messenger):
+    """
+    Handler for executing PyroSample statements in a more intuitive plate context.
+    """
+
+    def __init__(self, allowed_plates: Iterable[str] = ()):
+        self._inner_allowed_plates = frozenset(allowed_plates)
+
+    def __enter__(self):
+        self._plates = (
+            frozenset(p.name for p in pyro.poutine.runtime.get_plates())
+            | self._inner_allowed_plates
+        )
+        return super().__enter__()
+
+    def _is_local_plate(self, m: Messenger) -> bool:
+        return (
+            isinstance(m, pyro.poutine.plate_messenger.PlateMessenger)
+            and m.name not in self._plates
+        )
+
+    def _pyro_sample(self, msg) -> None:
+        if not msg["infer"].get("_original_pyrosample_dist", None):
+            return
+        msg["stop"] = True
+        msg["done"] = True
+        with pyro.poutine.messenger.block_messengers(
+            lambda m: m is self or self._is_local_plate(m)
+        ):
+            d = msg["infer"].pop("_original_pyrosample_dist")
+            msg["value"] = pyro.sample(
+                msg["name"],
+                d,
+                obs=msg["value"] if msg["is_observed"] else None,
+                infer=msg["infer"],
+            )
+
+
 def _make_name(prefix: str, name: str) -> str:
     return "{}.{}".format(prefix, name) if prefix else name
 
@@ -615,7 +659,13 @@ def __getattr__(self, name: str) -> Any:
                         value = (
                             pyro.deterministic(fullname, prior)
                             if isinstance(prior, torch.Tensor)
-                            else pyro.sample(fullname, prior)
+                            else pyro.sample(
+                                fullname,
+                                prior,
+                                infer=_PyroSampleInferDict(
+                                    _original_pyrosample_dist=prior
+                                ),
+                            )
                         )
                         context.set(fullname, value)
                     return value
diff --git a/tests/nn/test_module.py b/tests/nn/test_module.py
index 07c4daedd1..665a63d5a5 100644
--- a/tests/nn/test_module.py
+++ b/tests/nn/test_module.py
@@ -1084,3 +1084,46 @@ def forward(self):
     with pyro.settings.context(module_local_params=use_module_local_params):
         model = Model()
         pyro.render_model(model)
+
+
+def test_pyrosample_platescope():
+
+    class Model(pyro.nn.PyroModule):
+        def __init__(self, num_inputs, num_outputs):
+            super().__init__()
+            self.num_inputs = num_inputs
+            self.num_outputs = num_outputs
+            self.linear = pyro.nn.PyroModule[torch.nn.Linear](num_inputs, num_outputs)
+            self.linear.weight = pyro.nn.PyroSample(
+                dist.Normal(0, 1).expand([num_outputs, num_inputs]).to_event(2)
+            )
+            self.linear.bias = pyro.nn.PyroSample(
+                dist.Normal(0, 1).expand([num_outputs]).to_event(1)
+            )
+
+        @pyro.nn.PyroSample
+        def scale(self):
+            return (
+                pyro.distributions.LogNormal(0, 1)
+                .expand([self.num_outputs])
+                .to_event(1)
+            )
+
+        @pyro.nn.PyroSamplePlateScope()
+        def forward(self, x):
+            with pyro.plate("data", x.shape[-2], dim=-1):
+                assert (
+                    len(self.linear.weight.shape) == 2
+                    or self.linear.weight.shape[-3] != 1
+                )  # sampled outside data plate
+                loc = self.linear(x)
+                assert (
+                    len(self.scale.shape) == 1 or self.scale.shape[-2] == 1
+                )  # sampled outside data plate
+                y = pyro.sample("y", dist.Normal(loc, self.scale).to_event(1))
+                assert y.shape[-2] == x.shape[-2]  # ordinary pyro.sample statement
+                return y
+
+    model = Model(3, 2)
+    x = torch.randn(4, 3)
+    assert model(x).shape == (4, 2)