diff --git a/src/ExcelReader.Core/Reader/XlsxReader.Enumerator.cs b/src/ExcelReader.Core/Reader/XlsxReader.Enumerator.cs
index f0f8c88..7600f3a 100644
--- a/src/ExcelReader.Core/Reader/XlsxReader.Enumerator.cs
+++ b/src/ExcelReader.Core/Reader/XlsxReader.Enumerator.cs
@@ -726,8 +726,21 @@ private static bool TryParseIsoDate(ReadOnlySpan<byte> utf8, out DateTime value)
                 {
                     chars[i] = (char)utf8[i];
                 }
-                return DateTime.TryParse(chars[..utf8.Length], CultureInfo.InvariantCulture,
-                    DateTimeStyles.RoundtripKind | DateTimeStyles.AllowWhiteSpaces, out value);
+                const DateTimeStyles dateParseFlag = DateTimeStyles.RoundtripKind | DateTimeStyles.AllowWhiteSpaces;
+                if (!DateTime.TryParse(chars[..utf8.Length], CultureInfo.InvariantCulture, dateParseFlag, out value))
+                {
+                    return false;
+                }
+                // DateTime spans years 1..9999 but ToOADate only accepts 0100-01-01 and later — it
+                // throws OverflowException ("Not a legal OleAut date") below that. A hand-written
+                // t="d" like "0024-02-29" parses fine and would have taken down the whole read, so
+                // treat anything with no serial representation as unparseable and keep it as text.
+                if (value.Year < 100)
+                {
+                    value = default;
+                    return false;
+                }
+                return true;
             }
 
             private static ReadOnlySpan<byte> ElementText(ReadOnlySpan<byte> inner, ReadOnlySpan<byte> openTag, ReadOnlySpan<byte> closeTag)
diff --git a/tests/ExcelReader.Fuzz/README.md b/tests/ExcelReader.Fuzz/README.md
index 389d94c..c28f8a3 100644
--- a/tests/ExcelReader.Fuzz/README.md
+++ b/tests/ExcelReader.Fuzz/README.md
@@ -92,6 +92,7 @@ the working corpus for both jobs.
 |---|---|---|
 | `xls-minifat-overflow.bin` | `xls` | `OverflowException` out of `XlsCompoundFile.ReadIntSectors`. `miniFatSectorCount` (16,777,215 in a 3.4 KB file) was the one header sector count not bounded against the container length, and `ReadIntSectors` multiplies it by `sectorSize` inside a `checked` block. Now rejected as `InvalidDataException`; regression test in `ExcelOpenAndOleErrorTests`. |
 | `xlsx-truncated-cellxfs.bin` | `xlsx` | `ArgumentOutOfRangeException` out of `XlsxReader.ParseStyleDateFlags`. A styles part truncated mid-`<cellXfs` open tag left the `'>'` search at -1, which then anchored the search for `</cellXfs>` before being checked. Now returns no date flags; the `IdxOf` helpers also treat a negative anchor as "not found". Regression test in `XlsxReaderTests`. |
+| `xlsx-isodate-year-below-100.bin` | `xlsx`, `xlsx-memory` | `OverflowException` ("Not a legal OleAut date") out of `XlsxReader.Enumerator.EmitIsoDate`. A `t="d"` cell holding `0024-02-29T21:00:00.000Z` parsed fine into a `DateTime` but has no OLE automation serial — `ToOADate` only accepts `0100-01-01` and later. Such values are now kept verbatim as text like any other unparseable `t="d"`. Regression test in `NamespacePrefixAndIsoDateTests`. |
 
 ## Seeds
 
diff --git a/tests/ExcelReader.Fuzz/corpus/xlsx-isodate-year-below-100.bin b/tests/ExcelReader.Fuzz/corpus/xlsx-isodate-year-below-100.bin
new file mode 100644
index 0000000..e7259bc
Binary files /dev/null and b/tests/ExcelReader.Fuzz/corpus/xlsx-isodate-year-below-100.bin differ
diff --git a/tests/ExcelReader.Tests/NamespacePrefixAndIsoDateTests.cs b/tests/ExcelReader.Tests/NamespacePrefixAndIsoDateTests.cs
index 6fa38c7..fb1aad1 100644
--- a/tests/ExcelReader.Tests/NamespacePrefixAndIsoDateTests.cs
+++ b/tests/ExcelReader.Tests/NamespacePrefixAndIsoDateTests.cs
@@ -159,6 +159,39 @@ public void UnparseableIsoDateCellIsKeptAsString()
             Assert.Equal("not-a-date", e.Current[0].GetString());
         }
 
+        // Regression: found by the xlsx-memory fuzz target. DateTime spans years 1..9999 but
+        // ToOADate only accepts 0100-01-01 and later, so a t="d" whose text parsed into an earlier
+        // year threw OverflowException ("Not a legal OleAut date") out of MoveNext. Such a value has
+        // no Excel serial at all, so it is kept verbatim as text like any other unparseable t="d".
+        [Theory]
+        [InlineData("0024-02-29T21:00:00.000Z")]
+        [InlineData("0001-01-01")]
+        [InlineData("0099-12-31T23:59:59")]
+        public void IsoDateBelowOaDateRangeIsKeptAsString(string text)
+        {
+            using MemoryStream ms = WorkbookBuilder.Build(
+                $"""<row r="1"><c r="A1" t="d"><v>{text}</v></c></row>""");
+            using XlsxReader reader = Excel.From(ms);
+            using XlsxReader.Enumerator e = reader.GetEnumerator();
+
+            Assert.True(e.MoveNext());
+            Assert.Equal(CellType.ExcelString, e.Current[0].Type);
+            Assert.Equal(text, e.Current[0].GetString());
+        }
+
+        // The first legal OADate value still round-trips as a date, so the guard is not off by a day.
+        [Fact]
+        public void IsoDateAtOaDateFloorIsStillParsedAsDate()
+        {
+            using MemoryStream ms = WorkbookBuilder.Build(
+                """<row r="1"><c r="A1" t="d"><v>0100-01-01</v></c></row>""");
+            using XlsxReader reader = Excel.From(ms);
+            using XlsxReader.Enumerator e = reader.GetEnumerator();
+
+            Assert.True(e.MoveNext());
+            Assert.Equal(CellType.Date, e.Current[0].Type);
+        }
+
         [Fact]
         public async Task IsoDateCellIsParsedAsDateAsync()
         {