diff --git a/tests/test_tinycss2.py b/tests/test_tinycss2.py
index fb1130a..44c7785 100644
--- a/tests/test_tinycss2.py
+++ b/tests/test_tinycss2.py
@@ -669,6 +669,24 @@ def test_serialize_rules_with_functions():
     assert serialize(rules) == source
 
 
+@pytest.mark.parametrize('source', [
+    r'1\65 5',    # unit "e5"
+    r'1\45 5',    # unit "E5"
+    r'2\65 3px',  # unit "e3px"
+])
+def test_serialize_dimension_e_unit(source):
+    # A dimension whose unit starts with e/E followed by a digit must not be
+    # serialized as a bare "1e5": that re-parses as a number in scientific
+    # notation, changing the token type (and value). Serialization has to keep
+    # escaping the leading e so the token round-trips as the same dimension.
+    (dimension,) = parse_component_value_list(source)
+    assert dimension.type == 'dimension'
+    (reparsed,) = parse_component_value_list(serialize([dimension]))
+    assert reparsed.type == 'dimension'
+    assert reparsed.value == dimension.value
+    assert reparsed.lower_unit == dimension.lower_unit
+
+
 def test_backslash_delim():
     source = '\\\nfoo'
     tokens = parse_component_value_list(source)
diff --git a/tinycss2/ast.py b/tinycss2/ast.py
index 2831abc..324955d 100644
--- a/tinycss2/ast.py
+++ b/tinycss2/ast.py
@@ -568,7 +568,7 @@ def _serialize_to(self, write):
         write(self.representation)
         # Disambiguate with scientific notation
         unit = self.unit
-        if unit in ('e', 'E') or unit.startswith(('e-', 'E-')):
+        if unit[0] in 'eE' and (len(unit) == 1 or unit[1] in '+-0123456789'):
             write('\\65 ')
             write(serialize_name(unit[1:]))
         else: