diff --git a/CHANGES.md b/CHANGES.md
index 60ba3ca37446..c194cf730785 100644
--- a/CHANGES.md
+++ b/CHANGES.md
@@ -70,6 +70,11 @@
 
 ## Breaking Changes
 
+* In Python and YAML, ReadFromJson now override the dtype from None to
+  an explicit False.  Most notably, string values like `"123"` are preserved
+  as strings rather than silently coerced (and possibly truncated) to numeric
+  values.  To retain the old behavior, pass `dtype=True` (or any other value
+  accepted by `pandas.read_json`).
 * X behavior was changed ([#X](https://github.com/apache/beam/issues/X)).
 
 ## Deprecations
diff --git a/sdks/python/apache_beam/io/textio.py b/sdks/python/apache_beam/io/textio.py
index 3de9709d7362..0d7803bcabb1 100644
--- a/sdks/python/apache_beam/io/textio.py
+++ b/sdks/python/apache_beam/io/textio.py
@@ -23,7 +23,9 @@
 from functools import partial
 from typing import TYPE_CHECKING
 from typing import Any
+from typing import Dict
 from typing import Optional
+from typing import Union
 
 from apache_beam import typehints
 from apache_beam.coders import coders
@@ -980,7 +982,12 @@ def WriteToCsv(
 
   @append_pandas_args(pandas.read_json, exclude=['path_or_buf'])
   def ReadFromJson(
-      path: str, *, orient: str = 'records', lines: bool = True, **kwargs):
+      path: str,
+      *,
+      orient: str = 'records',
+      lines: bool = True,
+      dtype: Union[bool, Dict[str, Any]] = False,
+      **kwargs):
     """A PTransform for reading json values from files into a PCollection.
 
     Args:
@@ -992,11 +999,14 @@ def ReadFromJson(
       lines (bool): Whether each line should be considered a separate record,
         as opposed to the entire file being a valid JSON object or list.
         Defaults to True (unlike Pandas).
+      dtype (bool): If True, infer dtypes; if a dict of column to dtype,
+        then use those; if False, then don’t infer dtypes at all.
+        Defaults to False (unlike Pandas).
       **kwargs: Extra arguments passed to `pandas.read_json` (see below).
     """
     from apache_beam.dataframe.io import ReadViaPandas
     return 'ReadFromJson' >> ReadViaPandas(
-        'json', path, orient=orient, lines=lines, **kwargs)
+        'json', path, orient=orient, lines=lines, dtype=dtype, **kwargs)
 
   @append_pandas_args(
       pandas.DataFrame.to_json, exclude=['path_or_buf', 'index'])
diff --git a/sdks/python/apache_beam/io/textio_test.py b/sdks/python/apache_beam/io/textio_test.py
index 1d852d171324..b134d615e20e 100644
--- a/sdks/python/apache_beam/io/textio_test.py
+++ b/sdks/python/apache_beam/io/textio_test.py
@@ -1743,6 +1743,36 @@ def test_json_read_write(self):
 
         assert_that(pcoll, equal_to(records))
 
+  def test_numeric_strings_preserved(self):
+    records = [
+        beam.Row(
+            as_string=str(ix),
+            as_float_string=str(float(ix)),
+            as_int=ix,
+            as_float=float(ix)) for ix in range(3)
+    ]
+    with tempfile.TemporaryDirectory() as dest:
+      with TestPipeline() as p:
+        # pylint: disable=expression-not-assigned
+        p | beam.Create(records) | beam.io.WriteToJson(
+            os.path.join(dest, 'out'))
+      with TestPipeline() as p:
+        pcoll = (
+            p
+            | beam.io.ReadFromJson(os.path.join(dest, 'out*'))
+            | beam.Map(lambda t: beam.Row(**dict(zip(type(t)._fields, t)))))
+
+        assert_that(pcoll, equal_to(records))
+
+        # This test should be redundant as Python equality does not equate
+        # numeric values with their string representations, but this is much
+        # more explicit about what we're asserting here.
+        def check_types(element):
+          for a, b in zip(element, records[0]):
+            assert type(a) == type(b), (a, b, type(a), type(b))
+
+        _ = pcoll | beam.Map(check_types)
+
 
 if __name__ == '__main__':
   logging.getLogger().setLevel(logging.INFO)