diff --git a/packages/markitdown/src/markitdown/converters/_plain_text_converter.py b/packages/markitdown/src/markitdown/converters/_plain_text_converter.py
index 6a26560a9..e6a28560e 100644
--- a/packages/markitdown/src/markitdown/converters/_plain_text_converter.py
+++ b/packages/markitdown/src/markitdown/converters/_plain_text_converter.py
@@ -52,9 +52,17 @@ def convert(
         stream_info: StreamInfo,
         **kwargs: Any,  # Options to pass to the converter
     ) -> DocumentConverterResult:
+        raw_content = file_stream.read()
         if stream_info.charset:
-            text_content = file_stream.read().decode(stream_info.charset)
+            try:
+                text_content = raw_content.decode(stream_info.charset)
+            except UnicodeDecodeError:
+                # The charset was guessed from a small sample of the content
+                # (e.g. the first 4k bytes) and may not hold for the whole
+                # file. Fall back to detecting from the full content.
+                text_content = str(from_bytes(raw_content).best())
         else:
+            text_content = str(from_bytes(raw_content).best())
             data = file_stream.read()
             detected = from_bytes(data).best()
             text_content = (
diff --git a/packages/markitdown/tests/test_module_misc.py b/packages/markitdown/tests/test_module_misc.py
index 18690d188..a2b73dd43 100644
--- a/packages/markitdown/tests/test_module_misc.py
+++ b/packages/markitdown/tests/test_module_misc.py
@@ -817,6 +817,16 @@ def test_input_as_strings() -> None:
     assert "# Test" in result.text_content
 
 
+def test_plain_text_charset_guessed_from_prefix() -> None:
+    # Charset is guessed from only the first 4096 bytes of the stream. If
+    # that prefix happens to be pure ASCII but a multi-byte UTF-8 character
+    # (e.g. an em dash) appears later, the charset would previously be
+    # mis-detected as "ascii" and decoding the full content would raise
+    # UnicodeDecodeError. See PlainTextConverter.convert.
+    markitdown = MarkItDown()
+    input_data = ("a" * 4096 + "em dash — end").encode("utf-8")
+    result = markitdown.convert_stream(io.BytesIO(input_data), file_extension=".txt")
+    assert "em dash — end" in result.text_content
 def _mock_response(content_disposition: str) -> MagicMock:
     response = MagicMock()
     response.headers = {"content-disposition": content_disposition}