diff --git a/tensorflow_privacy/privacy/analysis/dp_event.py b/tensorflow_privacy/privacy/analysis/dp_event.py
new file mode 100644
index 0000000..fbec1d1
--- /dev/null
+++ b/tensorflow_privacy/privacy/analysis/dp_event.py
@@ -0,0 +1,84 @@
+# Copyright 2021, The TensorFlow Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Standard DpEvent classes."""
+
+from typing import List
+
+import attr
+
+
+class DpEvent(object):
+  """Base class for `DpEvent`s.
+
+  A `DpEvent` describes a differentially private mechanism sufficiently for
+  computing the associated privacy losses, both in isolation and in combination
+  with other `DpEvent`s.
+  """
+
+
+@attr.s(frozen=True)
+class NoOpDpEvent(DpEvent):
+  """A `DpEvent` to represent operations with no privacy impact.
+
+  A `NoOpDpEvent` is generally never required, but it can be useful as a
+  placeholder where a `DpEvent` is expected, such as in tests or some live
+  accounting pipelines.
+  """
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class GaussianDpEvent(DpEvent):
+  """The Gaussian mechanism."""
+  noise_multiplier: float
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class SelfComposedDpEvent(DpEvent):
+  """A mechanism composed with itself multiple times."""
+  event: DpEvent
+  count: int
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class ComposedDpEvent(DpEvent):
+  """A series of composed mechanisms."""
+  events: List[SelfComposedDpEvent]
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class PoissonSampledDpEvent(DpEvent):
+  """An application of Poisson subsampling."""
+  sampling_probability: float
+  event: DpEvent
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class EqualBatchSampledDpEvent(DpEvent):
+  """An application of sampling exactly `batch_size` records."""
+  dataset_size: int
+  batch_size: int
+  event: DpEvent
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class ShuffledDatasetDpEvent(DpEvent):
+  """Shuffling a dataset and applying a mechanism to each partition."""
+  partition_events: ComposedDpEvent
+
+
+@attr.s(frozen=True, slots=True, auto_attribs=True)
+class TreeAggregationDpEvent(DpEvent):
+  """Applying a series of mechanisms with tree aggregation."""
+  round_events: ComposedDpEvent
+  max_record_occurences_across_all_rounds: int
diff --git a/tensorflow_privacy/privacy/analysis/dp_event_builder.py b/tensorflow_privacy/privacy/analysis/dp_event_builder.py
new file mode 100644
index 0000000..a0a9435
--- /dev/null
+++ b/tensorflow_privacy/privacy/analysis/dp_event_builder.py
@@ -0,0 +1,78 @@
+# Copyright 2021, The TensorFlow Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Builder class for ComposedDpEvent."""
+
+import collections
+
+from tensorflow_privacy.privacy.analysis import dp_event
+
+
+class DpEventBuilder(object):
+  """Constructs a `DpEvent` representing the composition of a series of events.
+
+  Two common use cases of the `DpEventBuilder` are 1) for producing and tracking
+  a ledger of `DpEvent`s during sequential accounting using a
+  `PrivacyAccountant`, and 2) for building up a description of a composite
+  mechanism for subsequent batch accounting.
+  """
+
+  def __init__(self):
+    self._events = collections.OrderedDict()
+    self._composed_event = None
+
+  def compose(self, event: dp_event.DpEvent, count: int = 1):
+    """Composes new event into event represented by builder.
+
+    Args:
+      event: The new event to compose.
+      count: The number of times to compose the event.
+    """
+    if not isinstance(event, dp_event.DpEvent):
+      raise TypeError('`event` must be a subclass of `DpEvent`. '
+                      f'Found {type(event)}.')
+    if not isinstance(count, int):
+      raise TypeError(f'`count` must be an integer. Found {type(count)}.')
+    if count < 1:
+      raise ValueError(f'`count` must be positive. Found {count}.')
+
+    if isinstance(event, dp_event.ComposedDpEvent):
+      for composed_event in event.events:
+        self.compose(composed_event, count)
+    elif isinstance(event, dp_event.SelfComposedDpEvent):
+      self.compose(event.event, count * event.count)
+    elif isinstance(event, dp_event.NoOpDpEvent):
+      return
+    else:
+      current_count = self._events.get(event, 0)
+      self._events[event] = current_count + count
+      self._composed_event = None
+
+  def build(self) -> dp_event.DpEvent:
+    """Builds and returns the composed DpEvent represented by the builder."""
+    if not self._composed_event:
+      self_composed_events = []
+      for event, num_self_compositions in self._events.items():
+        if num_self_compositions == 1:
+          self_composed_events.append(event)
+        else:
+          self_composed_events.append(
+              dp_event.SelfComposedDpEvent(event, num_self_compositions))
+      if not self_composed_events:
+        return dp_event.NoOpDpEvent()
+      elif len(self_composed_events) == 1:
+        self._composed_event = self_composed_events[0]
+      else:
+        self._composed_event = dp_event.ComposedDpEvent(self_composed_events)
+
+    return self._composed_event
diff --git a/tensorflow_privacy/privacy/analysis/dp_event_builder_test.py b/tensorflow_privacy/privacy/analysis/dp_event_builder_test.py
new file mode 100644
index 0000000..a10d4bb
--- /dev/null
+++ b/tensorflow_privacy/privacy/analysis/dp_event_builder_test.py
@@ -0,0 +1,73 @@
+# Copyright 2021, The TensorFlow Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Tests for DpEventBuilder."""
+
+from absl.testing import absltest
+from tensorflow_privacy.privacy.analysis import dp_event
+from tensorflow_privacy.privacy.analysis import dp_event_builder
+
+_gaussian_event = dp_event.GaussianDpEvent(1.0)
+_poisson_event = dp_event.PoissonSampledDpEvent(_gaussian_event, 0.1)
+_self_composed_event = dp_event.SelfComposedDpEvent(_gaussian_event, 3)
+_composed_event = dp_event.ComposedDpEvent(
+    [_self_composed_event, _poisson_event])
+
+
+class DpEventBuilderTest(absltest.TestCase):
+
+  def test_no_op(self):
+    builder = dp_event_builder.DpEventBuilder()
+    self.assertEqual(dp_event.NoOpDpEvent(), builder.build())
+
+  def test_single(self):
+    builder = dp_event_builder.DpEventBuilder()
+    builder.compose(_gaussian_event)
+    self.assertEqual(_gaussian_event, builder.build())
+
+  def test_compose_no_op(self):
+    builder = dp_event_builder.DpEventBuilder()
+    builder.compose(dp_event.NoOpDpEvent())
+    builder.compose(_gaussian_event)
+    builder.compose(dp_event.NoOpDpEvent())
+    self.assertEqual(_gaussian_event, builder.build())
+
+  def test_compose_self(self):
+    builder = dp_event_builder.DpEventBuilder()
+    builder.compose(_gaussian_event)
+    builder.compose(_gaussian_event, 2)
+    self.assertEqual(_self_composed_event, builder.build())
+
+  def test_compose_heterogenous(self):
+    builder = dp_event_builder.DpEventBuilder()
+    builder.compose(_gaussian_event)
+    builder.compose(_poisson_event)
+    builder.compose(_gaussian_event, 2)
+    self.assertEqual(_composed_event, builder.build())
+
+  def test_compose_complex(self):
+    builder = dp_event_builder.DpEventBuilder()
+    builder.compose(_gaussian_event, 2)
+    builder.compose(_composed_event)
+    builder.compose(_poisson_event)
+    builder.compose(_composed_event, 2)
+
+    expected_event = dp_event.ComposedDpEvent([
+        dp_event.SelfComposedDpEvent(_gaussian_event, 11),
+        dp_event.SelfComposedDpEvent(_poisson_event, 4)
+    ])
+    self.assertEqual(expected_event, builder.build())
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/tensorflow_privacy/privacy/analysis/privacy_accountant.py b/tensorflow_privacy/privacy/analysis/privacy_accountant.py
new file mode 100644
index 0000000..2f1265f
--- /dev/null
+++ b/tensorflow_privacy/privacy/analysis/privacy_accountant.py
@@ -0,0 +1,101 @@
+# Copyright 2021, The TensorFlow Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""PrivacyAccountant abstract base class."""
+
+import abc
+
+from tensorflow_privacy.privacy.dp_event import dp_event
+from tensorflow_privacy.privacy.dp_event import dp_event_builder
+
+
+class PrivacyAccountant(metaclass=abc.ABCMeta):
+  """Abstract base class for privacy accountants."""
+
+  def __init__(self):
+    self._ledger = dp_event_builder.DpEventBuilder()
+
+  @abc.abstractmethod
+  def is_supported(self, event: dp_event.DpEvent) -> bool:
+    """Checks whether the `DpEvent` can be processed by this accountant.
+
+    In general this will require recursively checking the structure of the
+    `DpEvent`. In particular `ComposedDpEvent` and `SelfComposedDpEvent` should
+    be recursively examined.
+
+    Args:
+      event: The `DpEvent` to check.
+
+    Returns:
+      True iff this accountant supports processing `event`.
+    """
+
+  @abc.abstractmethod
+  def _compose(self, event: dp_event.DpEvent, count: int = 1):
+    """Update internal state to account for application of a `DpEvent`.
+
+    Calls to `get_epsilon` or `get_delta` after calling `_compose` will return
+    values that account for this `DpEvent`.
+
+    Args:
+      event: A `DpEvent` to process.
+      count: The number of times to compose the event.
+    """
+
+  def compose(self, event: dp_event.DpEvent, count: int = 1):
+    """Update internal state to account for application of a `DpEvent`.
+
+    Calls to `get_epsilon` or `get_delta` after calling `compose` will return
+    values that account for this `DpEvent`.
+
+    Args:
+      event: A `DpEvent` to process.
+      count: The number of times to compose the event.
+
+    Raises:
+      TypeError: `event` is not supported by this `PrivacyAccountant`.
+    """
+    if not self.is_supported(event):
+      raise TypeError(f'`DpEvent` {event} is of unsupported type.')
+    self._ledger.compose(event, count)
+    self._compose(event, count)
+
+  @property
+  def ledger(self) -> dp_event.DpEvent:
+    """Returns the (composed) `DpEvent` processed so far by this accountant."""
+    return self._ledger.build()
+
+  @abc.abstractmethod
+  def get_epsilon(self, target_delta: float) -> float:
+    """Gets the current epsilon.
+
+    Args:
+      target_delta: The target delta.
+
+    Returns:
+      The current epsilon, accounting for all composed `DpEvent`s.
+    """
+
+  def get_delta(self, target_epsilon: float) -> float:
+    """Gets the current delta.
+
+    An implementer of `PrivacyAccountant` may choose not to override this, in
+    which case `NotImplementedError` will be raised.
+
+    Args:
+      target_epsilon: The target epsilon.
+
+    Returns:
+      The current delta, accounting for all composed `DpEvent`s.
+    """
+    raise NotImplementedError()