I spent a day on China's paid annual leave rules, and the interesting part was not the day counts. It was that two of the questions people actually argue about are not answered in the regulation at all. They are answered in a 2009 letter from a ministry's general office, and restated in a city bureau's reply to a resident published last November.

The site version of this is here: China annual leave: 5, 10 or 15 days, and the 12-month gate. This post is about the stack of documents and the three places where a careful reading beats the summary everyone repeats.

The stack, in order of authority

Layer Document Year
Regulation State Council Order No. 514, in force 1 January 2008 2007
Implementing rules Ministry of Human Resources and Social Security Order No. 1 2008
Interpretation General office letter, Ren She Ting Han [2009] No. 149 2009
Local answer A city human resources bureau reply to a resident question 2025

The regulation gives three bands: 5 working days for 1 to under 10 years of cumulative work, 10 for 10 to under 20, 15 from 20. Public holidays and weekly rest days sit outside the count. Everything else that matters is in the lower layers.

Trap 1: "continuous 12 months" does not mean the 12 months before this job

Article 2 of the regulation grants leave to an employee who has worked continuously for a year. Every summary I read online turns that into "you must work 12 months at your current employer before you get any leave", which is not what the text says and not what the ministry letter says.

The letter settles it: the 12 continuous months can be at the same employer or across different employers. In practice the test is whether the person has ever had a continuous 12-month stretch of work. A hire with eight years of service behind them qualifies in week one. A fresh graduate does not qualify until 12 unbroken months have passed, and then gets a fraction of the year, not the full entitlement.

Trap 2: pro-rata applies to changing employer, not to changing band

The implementing rules pro-rate the first year: days left in the calendar year, divided by 365, times the full-year entitlement, part-days dropped. I assumed the same logic applied when someone's cumulative service crosses 10 years in July: five days before the anniversary, ten after.

The city answer published in November 2025 says otherwise. Crossing a band mid-year is settled at the higher number for the whole year. That answer is one bureau's, not a sentence in the regulation, so I flagged it as such rather than presenting it as national law. But it is the only written official answer I could find, and it runs against the intuitive reading.

Trap 3: the 300% is not a bonus on top of your wage

Unused leave is paid at 300% of the daily wage, and the text says that figure includes the normal wage for the day. So the premium is 200%, and two thirds of the payment is compensation while one third is wages the employee would have received anyway.

The base has two reductions people miss: it is the average monthly wage of the previous 12 months, and overtime pay is stripped out before dividing by 21.75. That 21.75 is a national figure from a 2008 notice, calculated as (365 − 104) ÷ 12.

The arithmetic, as assertions

I write these checks as assertions so the numbers in the article cannot drift from the formulas:

from datetime import date import math def entitlement(years): if years < 1: return 0 if years < 10: return 5 if years < 20: return 10 return 15 def pro_rata_join(start, year, years): days_left = (date(year, 12, 31) - start).days + 1 return math.floor(days_left / 365 * entitlement(years)) # 1 July 2026: 184 days left, so 2 / 5 / 7 depending on the band assert pro_rata_join(date(2026, 7, 1), 2026, 3) == 2 assert pro_rata_join(date(2026, 7, 1), 2026, 12) == 5 assert pro_rata_join(date(2026, 7, 1), 2026, 25) == 7 # 1 December 2026: 31 days left, and the 15-day band still rounds to 1 assert pro_rata_join(date(2026, 12, 1), 2026, 3) == 0 assert pro_rata_join(date(2026, 12, 1), 2026, 25) == 1 def unused_pay(monthly, days): daily = monthly / 21.75 return round(daily * days * 3, 2), round(daily * days * 2, 2) assert unused_pay(12000, 5) == (8275.86, 5517.24) assert unused_pay(12000, 15) == (24827.59, 16551.72) def on_leaving(start, end, years, taken): elapsed = (end - start).days + 1 return max(0, math.floor(elapsed / 365 * entitlement(years)) - taken) assert on_leaving(date(2026, 1, 1), date(2026, 6, 30), 12, 2) == 2 

The 1 July row is the one I would have got wrong by hand: 184 ÷ 365 × 5 is 2.52, and the fraction is not carried. The three bands produce 2, 5 and 7 on the same start date, not 2.5, 5 and 7.5.

What the reading cost

About three hours, of which perhaps forty minutes was finding the letter. The ministry's own copy did not surface in search; the full text was reproduced by a provincial human resources department on a 2009 page. That is a pattern worth naming: when an interpretive letter is old, the best copy is often on a province's site, not the ministry's.

The 2025 answers were easier, and more useful than expected, because they answer the two questions the regulation leaves blank. They also show the shape of Chinese rule-following: the statute sets numbers, the ministry sets mechanics, and the city bureau answers what the clerk at the counter actually asks.

What I could not settle

Whether work for a foreign employer before arriving counts toward cumulative service. The 2009 letter lists Chinese employer types and periods counted under Chinese law, and does not name foreign employers. Whether the mid-year band rule holds outside the one city that answered. And whether unused days survive past the year without written consent, where the rules require consent but do not say what happens without it.

Drafted with an AI assistant from my own reading of the primary documents; every clause number and link here was checked against the source.