diff --git a/confidence/formats.py b/confidence/formats.py
index 7566f79..4c9a2fa 100644
--- a/confidence/formats.py
+++ b/confidence/formats.py
@@ -1,7 +1,7 @@
 import json
 import typing
 from abc import ABC, abstractmethod
-from collections.abc import Sequence
+from collections.abc import Callable, Sequence
 from dataclasses import dataclass, replace
 from os import PathLike
 from pathlib import Path
@@ -26,6 +26,7 @@ class Format(ABC):
 
     suffix: str = ''  #: the default file path suffix for a configuration file of this Format
     encoding: str = 'utf-8'  #: the default text encoding for reading from binary I/O
+    value_fallback: Callable[[str], typing.Any] = str  #: the fallback 'factory' for unparseable single values
 
     def load(self, fp: typing.TextIO) -> typing.Any:
         return self.loads(fp.read())
@@ -38,6 +39,14 @@ def loadf(self, fpath: str | PathLike, encoding: str | None = None) -> typing.An
         with Path(fpath).open('rt', encoding=encoding or self.encoding) as fp:
             return self.load(fp)
 
+    def loadv(self, string: str) -> typing.Any:
+        try:
+            # hope the format implementation will be able to read string as a value formatted value
+            return self.loads(string)
+        except ValueError:
+            # use the fallback otherwise
+            return self.value_fallback(string)
+
     def dump(self, value: typing.Any, fp: typing.TextIO) -> None:
         fp.write(self.dumps(value))
 
diff --git a/confidence/io.py b/confidence/io.py
index f10a1b6..a6f4abe 100644
--- a/confidence/io.py
+++ b/confidence/io.py
@@ -101,8 +101,8 @@ def dotted(name: str) -> str:
     # include the number of variables matched for debugging purposes
     LOG.info(f'reading configuration from {len(values)} {prefix}* environment variables')
 
-    # pass value to yaml.safe_load to align data type transformation with reading values from files
-    return Configuration({dotted(name): format.loads(value) for name, value in values.items()})
+    # pass value to format.loadv to align data type transformation with reading values from files
+    return Configuration({dotted(name): format.loadv(value) for name, value in values.items()})
 
 
 def read_envvar_file(name: str, format: Format = YAML) -> Configuration:
diff --git a/tests/test_formats.py b/tests/test_formats.py
index 85f19f5..fc5507a 100644
--- a/tests/test_formats.py
+++ b/tests/test_formats.py
@@ -7,7 +7,42 @@
 from confidence.formats import JSON, TOML, YAML
 
 
-@pytest.mark.parametrize('format', (JSON, TOML, YAML))
+@pytest.mark.parametrize(
+    'format',
+    (
+        pytest.param(JSON, id='json'),
+        pytest.param(TOML, id='toml'),
+        pytest.param(YAML, id='yaml'),
+    ),
+)
+@pytest.mark.parametrize(
+    ('string', 'value'),
+    [
+        ('null', None),
+        ('true', True),
+        ('1', 1),
+        ('42.0', 42.0),
+        ('a string', 'a string'),
+        ("single'quote", "single'quote"),
+        ('double"quote', 'double"quote'),
+    ],
+)
+def test_singular_value_from_string(format, string, value):
+    if (format, value) == (TOML, None):
+        # None / null / nil is not supported by the TOML spec, see https://github.com/toml-lang/toml/issues/30
+        pytest.skip('None is unsupported for TOML format')
+
+    assert format.loadv(string) == value
+
+
+@pytest.mark.parametrize(
+    'format',
+    (
+        pytest.param(JSON, id='json'),
+        pytest.param(TOML, id='toml'),
+        pytest.param(YAML, id='yaml'),
+    ),
+)
 @pytest.mark.parametrize('value', (None, True, 1, 42.0, 'a string'))
 def test_singular_value_roundtrip(format, value):
     if (format, value) == (TOML, None):
@@ -17,7 +52,15 @@ def test_singular_value_roundtrip(format, value):
     assert format.loads(format.dumps(value)) == value
 
 
-@pytest.mark.parametrize('format', (JSON, TOML, YAML, YAML(suffix='.conf', encoding='utf-32')))
+@pytest.mark.parametrize(
+    'format',
+    (
+        pytest.param(JSON, id='json'),
+        pytest.param(TOML, id='toml'),
+        pytest.param(YAML, id='yaml'),
+        pytest.param(YAML(suffix='.conf', encoding='utf-32'), id='yaml-conf-32'),
+    ),
+)
 @pytest.mark.parametrize(
     'value',
     (
diff --git a/tests/test_io.py b/tests/test_io.py
index 8de5db9..c26f347 100644
--- a/tests/test_io.py
+++ b/tests/test_io.py
@@ -370,6 +370,35 @@ def test_load_name_envvars():
         assert not load_name('foo', 'bar', load_order=(read_envvars,))
 
 
+@pytest.mark.parametrize(
+    'format',
+    (
+        pytest.param(JSON, id='json'),
+        pytest.param(TOML, id='toml'),
+        pytest.param(YAML, id='yaml'),
+    ),
+)
+def test_load_name_envvars_value_types(format):
+    env = {
+        'FOO_KEY': 'foo',
+        'FOO_TYPES_NUM': '42',
+        'FOO_TYPES_FLT': '42.0',
+        'FOO_TYPES_BOL': 'true',
+        'FOO_TYPES_ST1': 'str',
+        'FOO_TYPES_ST2': '"str"',
+    }
+
+    with patch('confidence.io.environ', env):
+        config = load_name('foo', load_order=(read_envvars,), format=format)
+
+    assert config.key == 'foo'
+    assert config.types.num == 42
+    assert config.types.flt == 42.0
+    assert config.types.bol == True  # noqa: E712 (tomlkit uses a customized type that not is True but == True)
+    assert config.types.st1 == 'str'
+    assert config.types.st2 == 'str'
+
+
 def test_load_name_envvar_file(test_files):
     env = {
         'FOO_CONFIG_FILE': path.join(test_files, 'foo.yaml'),