Roman numeral conventions: where the rules actually live

Roman numerals use seven symbols: I = 1,V = 5, X = 10,L = 50, C = 100,D = 500, M= 1000. The modern reading rules are well-known. What’s less well-known is that those rules are mostly retroactive — the Romans themselves were inconsistent, and several modern domains still flout the “standard” rules on purpose.

The standard modern reading rules

1. Symbols generally add left-to-right.MCC = 1000 + 100 + 100 = 1200.
2. A smaller symbol before a larger one subtracts.IV = 4 (5 − 1), not IIII.IX = 9, XL = 40, XC = 90,CD = 400, CM = 900.
3. Only I, X, and C subtract. NeverVL (45) — write XLV. NeverLC (50) — write L... wait, that’s 50. Point is, V, L, D never subtract.
4. A subtracting symbol can only precede the next two larger.I before V or X only. X before L or C only.C before D or M only.
5. No symbol repeats more than three times in a row.III is fine. IIII is not (per modern convention) — write IV instead.

Where the “rules” come from

Ancient Roman inscriptions routinely use forms that violate modern rules. Roman coinage of the Republic era wrote 4 as IIII; subtractive notation (IV) appears later and was never universal. The Colosseum’s gate numbers run I,II, III, IIII,V — not IV.

The rule set we teach today was largely standardised in the 19th century by typographers and schoolbook authors who wanted a single consistent system. It is the most common modern usage — but if you find a 1300s manuscript writing IIII or VIIII, it’s not wrong by the standards of its own time.

Where the rules are still broken on purpose

Clock faces

Most analog clocks with Roman numerals use IIIIfor 4, not IV. Reasons offered range from visual balance with the VIIIon the opposite side, to Louis XIV’s alleged annoyance atIV looking like a partial IVPITER(Jupiter), to simple convention. Whatever the cause, the practice is so universal that IV on a clock face looks wrong.

Film and TV credits

Hollywood traditionally uses Roman numerals for copyright years — partly tradition, partly to make the year less immediately legible. Modern usage is properly subtractive (MMXXIV, not MMXXIIII).

Monarch and pope numbering

Roman numerals are the convention for monarchs (Elizabeth II, Louis XIV) and popes (Benedict XVI). Subtractive notation applies normally — Edward IV, Henry IX (a name no English monarch has actually held, but you get the idea).

Manuscript and book pagination

Front matter (preface, table of contents, list of figures) is conventionally paginated in lowercase Roman numerals (i, ii, iii, iv, v, …). The main text starts over at Arabic 1. Most scholarly publishers and academic theses still use this convention.

The numbers Roman numerals can’t express

• Zero.Roman numerals have no symbol for zero. The number didn’t exist as a concept in standard Roman mathematics.
• Fractions below 1/12.Romans used a duodecimal subdivision (uncia = 1/12) with named fractions. Decimals don’t fit the system.
• Numbers above ~3,999 cleanly. The traditional symbols stop at M = 1000. For larger numbers, several extensions exist — a bar over the symbol (vinculum) multiplied it by 1000, so V̄ = 5000. But no single unambiguous standard exists.
• Negative numbers. No notation.

A worked conversion: 2026

Decompose: 2000 + 0 + 20 + 6.

• 2000 = MM
• 0 (hundreds) = (nothing)
• 20 = XX
• 6 = VI

Concatenate: MMXXVI. Read it back to verify: 1000 + 1000 + 10 + 10 + 5 + 1 = 2026. ✓

The pragmatic takeaway

For copyright lines, monarch names, and most modern uses: follow the standard rules above. For clock faces, writeIIII. For ancient inscriptions, document what’s actually there — don’t “correct” a 1500-year-old engraving to match a 19th-century textbook.

Sources: Donald E. Knuth, The Art of Computer Programming, Vol. 2 (3rd ed., §4.1); J. C. Reynolds,Numbers and Calculations: A History of Notation(Springer, 2018); Vatican Library inscription catalogue (1992 ed.).

Step-by-step walkthrough: 1994 and 3999

Two more numbers that exercise different parts of the ruleset.

1994. Decompose into the four place-values: 1000 + 900 + 90 + 4.

• 1000 → M
• 900 → CM (subtractive: 1000 − 100)
• 90 → XC (subtractive: 100 − 10)
• 4 → IV (subtractive: 5 − 1)

Concatenate: MCMXCIV. Notice three of the four place-values use subtractive notation — 1994 happens to be the longest Hollywood copyright year using only subtractive groupings, which is why it appears in worked examples in every introductory textbook.

3999. Decompose: 3000 + 900 + 90 + 9.

• 3000 → MMM
• 900 → CM
• 90 → XC
• 9 → IX

Result: MMMCMXCIX — nine characters, the densest non-trivial Roman numeral inside the normally-expressible range. 4000 (MMMM) would violate the “no symbol repeats more than three times” rule under strict modern convention, which is why 3,999 is the de-facto upper bound for clean single-script Roman numerals.

For arbitrarily larger numbers you need the vinculum notation (a bar above the symbol multiplies it by 1,000). Unicode encodes the most common vinculum-multiplied forms at U+2180-U+2188, but support across fonts is patchy and most modern uses skip the concept entirely — see the Unicode L2 proposal for Ancient Roman Numerals (2007) for the encoding history.

Common mistakes when reading or writing Roman numerals

• Subtracting from the wrong target. ILdoesn’t mean 49. The subtracting symbol must be one of {I, X, C} and may only precede the next-two-larger symbols. 49 is XLIX (40 + 9), not IL.
• Stacking subtractions. IIX for 8 looks plausible but isn’t standard. Each subtractive group has exactly one smaller symbol before one larger. 8 is VIII.
• Using lowercase carelessly. Lowercase Roman numerals (i, ii, iii, iv, v) are conventional for book front-matter pagination but inappropriate for monarchs, popes, or copyright lines. Mixed-case (Iv) is always wrong.
• Treating digit groups independently. VCfor 95 ignores the “only I, X, C subtract” rule. The correct form is XCV. Similarly IC for 99 is a non-standard convenience used in stock-tickers and informal contexts — the formal form is XCIX.
• Assuming the Unicode codepoints normalise to the ASCII letters. U+2160 (Ⅰ), U+2161 (Ⅱ), etc., are dedicated Roman numeral codepoints that look like the ASCII letters but compare differently in string comparisons. Always normalise (NFKC) input before parsing.

When Roman numerals do NOT apply

• Anything requiring arithmetic. Adding, subtracting, multiplying, or dividing Roman numerals on paper is famously painful — which is one reason the Hindu-Arabic numerals replaced them across Europe between the 12th and 15th centuries. Convert to Arabic for any computation.
• Sort orders or version numbers in software. Lexical sorting ofI, II, III, IV, V, VI, VII, VIII, IX, X producesI, II, III, IV, IX, V, VI, VII, VIII, X — wrong by numeric order. Use Arabic numerals for anything a computer needs to sort.
• Negative numbers, zero, fractions, or any number above ~4,000.The notation simply doesn’t encode these cases cleanly. For ordinals and ranks under 100 you’re fine; for anything else, Arabic is the right tool.
• Multilingual contexts where the reader may not know the system. A copyright line of MMXXVI assumes the reader can decode it. Pair with the Arabic year if the audience is international or non-specialist.

For the underlying counting-and-notation history, see our math methodology page and the Unicode Number Forms block PDF (U+2150-U+218F) for the canonical codepoint reference.

The greedy algorithm in pseudocode

Every Roman-numeral converter on the web uses the same greedy algorithm — at each step, subtract the largest value whose symbol you can still write, append the symbol, and continue. The full lookup table including subtractive pairs:

pairs = [
  (1000, 'M'),  (900, 'CM'), (500, 'D'),  (400, 'CD'),
  (100,  'C'),  (90,  'XC'), (50,  'L'),  (40,  'XL'),
  (10,   'X'),  (9,   'IX'), (5,   'V'),  (4,   'IV'),
  (1,    'I'),
]
result = ''
n = input_number
for (value, symbol) in pairs:
  while n >= value:
    result += symbol
    n -= value
return result

For 1994: subtract 1000 (M), subtract 900 (CM), subtract 90 (XC), subtract 4 (IV). Result: MCMXCIV. The algorithm is O(13) per number — constant time — and produces canonical output for every input in the standard 1-3999 range. Reverse-direction (parse a Roman numeral to integer) is similar: scan left to right, add each value, subtract twice when a smaller precedes a larger.

Number-to-numeral quick reference (1-100)

For quick lookups without going through the place-value decomposition every time:

Notice the subtractive groupings only appear at the boundaries (4, 9, 40, 90, 400, 900). Every other number is additive. This is the underlying reason the standard algorithm (greedy: subtract the largest possible value at each step) always produces the canonical form.

Special-purpose ranges to memorise:

• Birth years and copyright lines: 1900 = MCM, 2000 = MM, 2026 = MMXVI... wait, 2026 = MMXXVI (the example we worked above).
• Movie sequels: II through IX cover almost every franchise; X (Rocky X, Fast X) starts appearing in the 2020s.
• Super Bowls: Super Bowl LX is February 2026 (year LX = 60).
• Monarchs: Elizabeth II ruled 1952-2022; Louis XIV ruled 1643-1715 (72 years, the longest verified European monarch reign).