diff --git a/src/py_avro_schema/_schemas.py b/src/py_avro_schema/_schemas.py
index 884c1f9..9f59742 100644
--- a/src/py_avro_schema/_schemas.py
+++ b/src/py_avro_schema/_schemas.py
@@ -87,7 +87,17 @@ class TDMissingMarker(str):
     ...
 
 
+class BytesTDMissingMarker(bytes):
+    """
+    Similar to `TDMissingMarker` above, but is required for Typed Dict that have non-required values in `bytes`, as Avro
+    is unable to manage Unions of `string` and `bytes`.
+    """
+
+    ...
+
+
 TD_MISSING_MARKER = TDMissingMarker("__td_missing__")
+BYTES_TD_MISSING_MARKER = BytesTDMissingMarker(b"__td_missing__")
 
 
 class TypeNotSupportedError(TypeError):
@@ -885,6 +895,9 @@ def _validate_union(args: tuple[Any, ...]) -> None:
         :return: None
         :raises: TypeError if the Union types are invalid
         """
+        if str in args and bytes in args:
+            raise TypeError("Avro does not support Union of types bytes and string")
+
         if type(None) not in args and TDMissingMarker not in args:
             if any(
                 # Enum is treated as a Sequence
@@ -1146,7 +1159,8 @@ def __init__(
         if self.default != dataclasses.MISSING:
             if isinstance(self.schema, UnionSchema):
                 self.schema.sort_item_schemas(self.default)
-            typeguard.check_type("default_value", self.default, self.py_type)
+            if self.default != TD_MISSING_MARKER:
+                typeguard.check_type("default_value", self.default, self.py_type)
         else:
             if Option.DEFAULTS_MANDATORY in self.options:
                 raise TypeError(f"Default value for field {self} is missing")
@@ -1387,21 +1401,25 @@ def _record_field(self, py_field: tuple[str, Type]) -> RecordField:
         """Return an Avro record field object for a given TypedDict field"""
         aliases, actual_type = get_field_aliases_and_actual_type(py_field[1])
 
+        # Avro does not handle Unions of bytes and string
+        marker_type = BytesTDMissingMarker if _is_bytes(actual_type) else TDMissingMarker
+
         default = dataclasses.MISSING
         if Option.MARK_NON_TOTAL_TYPED_DICTS in self.options and not self.is_total:
             # If a TypedDict is marked as total=False, it does not need to contain all the field. However, we need to
             # be able to distinguish between the fields that are missing from the ones that are present but set to None.
             # To do that, we extend the original type with str. We will later add a special string
             # (e.g., __td_missing__) as a marker at deserialization time.
-            actual_type = Union[actual_type, TDMissingMarker]  # type: ignore
+            actual_type = Union[actual_type, marker_type]  # type: ignore
             if _is_optional(actual_type):
                 # Note: this works since this schema does not implement `make_default` and the base implementation
                 # simply return the provided type (None in this case).
+                # We need to use the string TD_MISSING_MARKER as the schema cannot serialize bytes
                 default = TD_MISSING_MARKER  # type: ignore
         elif _is_not_required(actual_type):
             # A field can be marked with typing.NotRequired even in a TypedDict with is not marked with total=False.
             # Similarly as above, we extend the wrapped type with string.
-            actual_type = Union[_unwrap_not_required(actual_type), TDMissingMarker]  # type: ignore
+            actual_type = Union[_unwrap_not_required(actual_type), marker_type]  # type: ignore
 
         field_obj = RecordField(
             py_type=actual_type,
@@ -1433,6 +1451,14 @@ def _is_optional(py_type: Type) -> bool:
         return False
 
 
+def _is_bytes(py_type: Type) -> bool:
+    """Given a Union of types, checks if bytes is one of those"""
+    try:
+        return py_type is bytes or bytes in get_args(py_type)
+    except Exception:
+        return False
+
+
 def _is_not_required(py_type: Type) -> bool:
     """Checks if a type is marked with typing.NotRequired"""
     return get_origin(py_type) is NotRequired  # noqa
diff --git a/tests/test_primitives.py b/tests/test_primitives.py
index 5c96b3b..925a11f 100644
--- a/tests/test_primitives.py
+++ b/tests/test_primitives.py
@@ -430,6 +430,15 @@ def test_literal_different_types():
         py_avro_schema._schemas.schema(py_type)
 
 
+def test_union_bytes_string():
+    py_type = Union[str, bytes]
+    with pytest.raises(
+        TypeError,
+        match=re.escape("Avro does not support Union of types bytes and string"),
+    ):
+        py_avro_schema._schemas.schema(py_type)
+
+
 def test_optional_str():
     py_type = Optional[str]
     expected = ["string", "null"]
diff --git a/tests/test_typed_dict.py b/tests/test_typed_dict.py
index 8434108..3ce9b73 100644
--- a/tests/test_typed_dict.py
+++ b/tests/test_typed_dict.py
@@ -116,6 +116,8 @@ class PyType(TypedDict, total=False):
         age: int | None
         invalid: InvalidEnumSymbol | None
         valid: ValidEnumSymbol | None
+        bytes_data: bytes
+        bytes_data_nullable: bytes | None
 
     expected = {
         "fields": [
@@ -144,6 +146,11 @@ class PyType(TypedDict, total=False):
                     "null",
                 ],
             },
+            {
+                "name": "bytes_data",
+                "type": "bytes",
+            },
+            {"default": "__td_missing__", "name": "bytes_data_nullable", "type": ["bytes", "null"]},
         ],
         "name": "PyType",
         "type": "record",