diff --git a/lib/fixRanges.spec.ts b/lib/fixRanges.spec.ts
index e5c01c6..abde247 100644
--- a/lib/fixRanges.spec.ts
+++ b/lib/fixRanges.spec.ts
@@ -90,8 +90,6 @@ describe('fixRanges prefixes', () => {
     isFixed('=SUM(A:C!A1)', "=SUM('A:C'!A1)");
     isFixed('=SUM(B:C!A1)', "=SUM('B:C'!A1)");
     isFixed('=SUM(C:D!A1)', "=SUM('C:D'!A1)");
-    isFixed('=SUM(A:8!A1)', "=SUM('A:8'!A1)");
-    isFixed('=SUM(B:8!A1)', "=SUM('B:8'!A1)");
     isFixed('=SUM(8:D!A1)', "=SUM('8:D'!A1)");
     isFixed('=SUM(10:23!A1)', "=SUM('10:23'!A1)");
     isFixed('=A:B!A1', '=A:B!A1');
@@ -126,12 +124,27 @@ describe('fixRanges prefixes', () => {
   });
 
   test('a digit-leading second sheet name goes to the range operator', () => {
-    isFixed('=SUM(Sheet1:1!A1)', "=SUM('Sheet1:1'!A1)");
-    isFixed('=SUM(X:1!A1)', "=SUM('X:1'!A1)");
+    // Excel fixes Sheet1:1!A1 by quoting the RHS, choosing the latter of these interpretations:
+    // - 'Sheet1:1'!A1 is a 3-D reference spanning the sheets Sheet1 to 1
+    // - Sheet1:'1'!A1 is a range from (misleadingly named) _defined name_ Sheet1 to cell '1'!A1
+    isFixed('=SUM(Sheet1:1!A1)', "=SUM(Sheet1:'1'!A1)");
+    isFixed('=SUM(X:1!A1)', "=SUM(X:'1'!A1)");
     isFixed("=SUM(Sheet1:'1'!A1)", "=SUM(Sheet1:'1'!A1)");
     isFixed("=SUM(Jan:'2020plan'!A1)", "=SUM(Jan:'2020plan'!A1)");
+    isFixed('=SUM([Book.xlsx]Alpha:3!A1)', "=SUM([Book.xlsx]Alpha:'3'!A1)");
+    // A number cannot be the left operand of the range operator, so these are sheet ranges.
+    isFixed('=SUM(1:5!A1)', "=SUM('1:5'!A1)");
     isFixed("=SUM('1:5'!A1)", "=SUM('1:5'!A1)");
     isFixed('=SUM([Book.xlsx]1:3!A1)', "=SUM('[Book.xlsx]1:3'!A1)");
+    isFixed('=SUM(Sheet1:1!A1)', "=SUM(Sheet1:'1'!A1)", { xlsx: true });
+    isFixed('=SUM(1:5!A1)', "=SUM('1:5'!A1)", { xlsx: true });
+    // Quoting the whole prefix makes a sheet range, regardless of what the right name looks like.
+    isFixed("=SUM('Sheet1:1'!A1)", "=SUM('Sheet1:1'!A1)");
+  });
+
+  test.fails('a digit-leading second sheet name goes to the range operator even when not all digits', () => {
+    // Fails today because 2020plan is lexed as the number 2020 followed by the name plan.
+    isFixed('=SUM(Jan:2020plan!A1)', "=SUM(Jan:'2020plan'!A1)");
   });
 
   test('a left side that is also a cell address wins over a 3D reference', () => {
diff --git a/lib/mergeRefTokens.ts b/lib/mergeRefTokens.ts
index bd211db..62f6fd6 100644
--- a/lib/mergeRefTokens.ts
+++ b/lib/mergeRefTokens.ts
@@ -7,6 +7,18 @@ const BR_OPEN = 91; // [
 const BR_CLOSE = 93; // ]
 const COLON = 58;
 
+// In an unquoted pair of sheet names, a digit-led right name makes the colon a range operator
+// rather than marking a sheet range, unless the left name is a number, which cannot be the range
+// operator's left operand.
+const reDigitLed = /^\d/;
+const reAllDigits = /^\d+$/;
+
+// The sheet name in an unquoted prefix: what follows its [workbook] and any path before that.
+function sheetNameOf (unquotedPrefix: string): string {
+  const close = unquotedPrefix.lastIndexOf(']');
+  return close === -1 ? unquotedPrefix : unquotedPrefix.slice(close + 1);
+}
+
 const validRunsMerge = [
   // A1 | A1:B2 | A:B | 1:2 | A1:B
   [ REF_CELL, ':', REF_CELL ],
@@ -93,6 +105,7 @@ const matcher = (
   let cols = 0;
   let brackets = 0;
   let inPrefix = false;
+  let rightSheetName: string | undefined;
   // the longest run so far that ended on a terminal: a longer run may turn out
   // to be invalid, in which case we fall back to the best valid subset run
   let best = 0;
@@ -109,6 +122,24 @@ const matcher = (
     const value = token.value;
     if (inPrefix) {
       const firstChar = value.charCodeAt(0);
+      if (
+        firstChar !== COLON &&
+        (token.type === CONTEXT || token.type === CONTEXT_QUOTE || token.type === REF_NAMED)
+      ) {
+        const name = sheetNameOf(token.type === CONTEXT_QUOTE ? unquotePrefix(value) : value);
+        if (cols === 0) {
+          rightSheetName = name;
+        }
+        // This is entered only if an earlier token set cols to 1, so never for a quoted sheet span
+        else if (
+          cols === 1 &&
+          rightSheetName != null &&
+          reDigitLed.test(rightSheetName) &&
+          !reAllDigits.test(name)
+        ) {
+          break runloop;
+        }
+      }
       if (firstChar === COLON) {
         cols++;
         if (cols >= 2 || brackets) { break runloop; }
diff --git a/lib/parseA1Ref.spec.ts b/lib/parseA1Ref.spec.ts
index 75fde98..579e940 100644
--- a/lib/parseA1Ref.spec.ts
+++ b/lib/parseA1Ref.spec.ts
@@ -169,17 +169,28 @@ describe('parse A1 references', () => {
   });
 
   test('a period/digit-leading sheet names', () => {
-    // These aren't strictly legal, as sheets that begin with digits or periods must
-    // be quoted. But Excel will parse and convert them to quoted, so we do to.
+    // A sheet name that begins with a digit must be quoted. At the right end of an unquoted pair
+    // Excel quotes that name alone, which leaves the colon to the range operator.
     const range = { top: 0, left: 0, bottom: 0, right: 0 };
-    isA1Equal('Sheet1:1!A1', { context: [ 'Sheet1:1' ], range });
-    isA1Equal('Jan:2020plan!A1', { context: [ 'Jan:2020plan' ], range });
-    isA1Equal('X:1!A1', { context: [ 'X:1' ], range });
-    isA1Equal('[Book.xlsx]Alpha:3!A1', { context: [ 'Book.xlsx', 'Alpha:3' ], range });
+    // Unquoted, a digit-led second name makes the colon a range operator rather than a sheet range
+    // marker, so these are not a single reference.
+    expect(parseA1Ref('Sheet1:1!A1')).toBe(undefined);
+    expect(parseA1Ref('Jan:2020plan!A1')).toBe(undefined);
+    expect(parseA1Ref('X:1!A1')).toBe(undefined);
+    expect(parseA1Ref('[Book.xlsx]Alpha:3!A1')).toBe(undefined);
+    // A number cannot be the left operand of the range operator, so these are sheet ranges.
     isA1Equal('1:5!A1', { context: [ '1:5' ], range });
     isA1Equal('12:15!A1', { context: [ '12:15' ], range });
     isA1Equal('1:2020plan!A1', { context: [ '1:2020plan' ], range });
     isA1Equal('[Book.xlsx]1:3!A1', { context: [ 'Book.xlsx', '1:3' ], range });
+    // the same rule in the xlsx variant
+    isA1Equal('Sheet1:1!A1', undefined, { xlsx: true });
+    isA1Equal('Jan:2020plan!A1', undefined, { xlsx: true });
+    isA1Equal('[1]Alpha:3!A1', undefined, { xlsx: true });
+    isA1Equal('1:5!A1', { sheetName: '1:5', range }, { xlsx: true });
+    isA1Equal('[1]1:3!A1', { workbookName: '1', sheetName: '1:3', range }, { xlsx: true });
+    // Quoting the whole prefix makes a sheet range whatever the right name is.
+    isA1Equal("'Sheet1:1'!A1", { context: [ 'Sheet1:1' ], range });
   });
 
   test('a left side that is also a cell address wins over a sheet range', () => {
diff --git a/lib/parseRef.ts b/lib/parseRef.ts
index 9b4fb54..168ef59 100644
--- a/lib/parseRef.ts
+++ b/lib/parseRef.ts
@@ -210,6 +210,13 @@ const pExtendedContext: RefParserPart = (t, data, xlsx, r1c1, tokens) => {
     const d: Partial<RefParseDataCtx & RefParseDataXls> = {};
     const value = type === CONTEXT_QUOTE ? unquotePrefix(t.value) : t.value;
     splitContext(value, d, xlsx);
+    // A digit-led second sheet name makes the colon a range operator, not a sheet range marker,
+    // unless the first name is an integer, which cannot be the range operator's left operand.
+    const second = xlsx ? d.sheetName : d.context?.[0];
+    const first = xlsx ? data.sheetName : data.context?.at(-1);
+    if (second && /^\d/.test(second) && !/^\d+$/.test(first ?? '')) {
+      return;
+    }
     if (xlsx && d.sheetName && !d.workbookName) {
       data.sheetName += ':' + d.sheetName;
       return 1;