How to Find Your Real Market Salary Before You Negotiate

Start with the government data, not the salary sites

The best free source of American pay data is the Occupational Employment and Wage Statistics programme run by the Bureau of Labor Statistics. It is a mandatory-methodology survey of establishments rather than a collection of volunteered figures, it publishes the full wage distribution rather than a single average, and it breaks down by state, metropolitan area and industry. It has real weaknesses, covered below. It is still the anchor, and anything else you use should be checked against it.

The most recent release is the May 2025 estimates, published on 15 May 2026. It covers 155,495,730 jobs across roughly 830 occupations, and reports a mean annual wage across all occupations of 69,770 dollars and a median hourly wage of 24.51 dollars. Estimates are produced for the nation, for states and the District of Columbia, and for approximately 530 metropolitan and nonmetropolitan areas.

What percentiles mean, and why the median is the least useful number

The programme publishes wages at the 10th, 25th, 50th, 75th and 90th percentiles. The definition is mechanical: 10 percent of workers in that occupation earn at or below the 10th percentile figure, half earn at or below the median, and 10 percent earn at or above the 90th.

The median alone tells you almost nothing useful for a negotiation, because it hides the shape of the distribution. Two occupations with the same median can have completely different negotiating room. What you want is the gap between the 25th and 75th percentiles, which tells you how much of the pay in that job is determined by the job title and how much is determined by who you are and where you work. A narrow gap means the title sets the pay and negotiation moves little. A wide gap means there is real variation to argue about, and your job is to establish which part of it you sit in.

A worked example: reading a spread and setting a number

Take accountants and auditors, which the Bureau classifies under occupation code 13-2011. The May 2023 national estimates for this occupation, which are the most recent published as a single static national table with the full percentile set, were as follows.

Read that before you read any advice about it. The 25th to 75th band runs from 62,720 to 103,990 dollars, a spread of 41,270 dollars. The 90th percentile is roughly 2.7 times the 10th. The mean sits well above the median, which tells you the distribution has a long upper tail: a minority of people in this occupation earn a great deal more than typical, which drags the average up. If somebody quotes you the average accountant salary, they are quoting a number that describes almost nobody.

Geography moves this substantially. In the same May 2023 release, the annual mean wage for this occupation was 124,070 dollars in the San Jose-Sunnyvale-Santa Clara metropolitan area, 119,050 dollars in New York-Newark-Jersey City, 117,930 dollars in San Francisco-Oakland-Hayward, 104,850 dollars in Washington-Arlington-Alexandria and 102,050 dollars in Seattle-Tacoma-Bellevue. Those metro means are all above the national 75th percentile. If you work in one of those areas and you benchmark against the national median, you have talked yourself down before the conversation starts.

Now the currency check. In the May 2025 release, the same occupation shows employment of 1,449,500, a mean annual wage of 94,750 dollars and a median hourly wage of 40.23 dollars. The mean has moved from 90,780 to 94,750 dollars over two years, a rise of about 4.4 percent. That tells you two things: the percentile structure above is still roughly the right shape, and the levels are a little low, so an ask built purely on the older table is slightly stale.

A decision rule for setting the ask

Score yourself against four tests, each worth one point:

1. You are in a metropolitan area whose mean for this occupation is above the national mean.
2. You have more years in the occupation than a typical holder of the role you are applying for.
3. You are moving into a higher-paying industry for the same occupation, not a lower-paying one.
4. You hold a credential, licence or specialism that the posting explicitly asks for.

Zero or one point: anchor at the median. Two points: anchor midway between the median and the 75th percentile. Three points: anchor at the 75th percentile. Four points: anchor above the 75th percentile and be ready to justify it with specifics, not adjectives.

Applied to the table above, a candidate scoring three points anchors at 103,990 dollars rather than the median of 79,880, then adjusts upward for two years of wage growth, which puts the ask in the region of 108,000 dollars. That is a defensible number with a stated derivation, which is a different thing from a number you hoped for. Note also that this method produces an anchor, not a floor. Your floor is a separate decision about what you will actually accept.

What the survey does not count

This is where people get the arithmetic badly wrong. The Bureau defines the wages it collects as straight-time gross pay, exclusive of premium pay. Included are base rate, cost-of-living allowances, guaranteed pay, hazard pay, incentive pay such as commissions and production bonuses, tips and piece rates. Excluded are overtime pay, shift differentials, non-production bonuses, holiday bonuses, stock bonuses, meal and lodging allowances, and the employer cost of benefits such as health insurance and retirement contributions.

The consequences are specific. For an hourly job with heavy overtime, the survey understates what people actually take home. For a role with a large annual bonus or equity grant, the survey materially understates total compensation, sometimes by a very large margin, and comparing an equity-heavy offer against these figures will make the offer look better than it is. The survey also excludes self-employed workers entirely, so for occupations with substantial self-employment the published figures describe employees only.

The lag nobody mentions

These estimates are not a snapshot. According to the programme’s own technical note, the May 2025 estimates combine six semiannual panels collected over three years: May 2025, November 2024, May 2024, November 2023, May 2023 and November 2022. Wages collected in earlier panels are adjusted forward to the reference date using adjustment factors derived from the survey itself. The sample covers roughly 1.1 million establishments drawn from state unemployment insurance files, and the overall national response rate across the six panels was 66.2 percent by establishment.

So the data labelled May 2025 contains information gathered as far back as late 2022, adjusted forward, and was published in May 2026. In a stable occupation this barely matters. In an occupation where pay has moved sharply in the last two years, it matters a lot, and the published figure will trail the market. Adjust upward and say so rather than pretending the number is current.

Why crowd-sourced salary sites skew

The large salary comparison sites are useful for one thing the government data cannot do: they carry company names and specific internal job titles. That is genuinely valuable. Their weakness is structural and cannot be fixed by sample size.

• The sample selects itself. People submit their pay when they have a reason to. Those reasons are not evenly distributed across the workforce, and there is no weighting back to a known population, which is exactly what the government survey does.
• Nothing is verified. Entries are typed in by individuals, and there is usually no check against a payslip.
• Titles are blended. Two people with the same title in different companies can be doing different jobs at different levels, and the site averages them together.
• Entries age quietly. A figure submitted three years ago may still be sitting in the average with no visible date attached.
• Some sites tie access to submission. Requiring a submission to see the data shapes who submits.

Use them to learn the title structure and levelling at a specific employer. Do not use them as your primary evidence of what a job pays, and be sceptical of any figure presented to three significant digits with no sample size next to it.

Pay transparency laws are the free data source people ignore

A growing number of states require employers to disclose pay ranges, and this produces something the surveys cannot: current, employer-specific, role-specific ranges for jobs that are open right now.

The employment law firm Jackson Lewis, in its 2026 overview of pay transparency obligations, identifies states requiring pay range disclosure including California, Colorado, Hawaii, Illinois, Maryland, Massachusetts, Minnesota, New Jersey, New York, Vermont, Washington and the District of Columbia, with employer size thresholds varying by state, and Connecticut, Nevada and Rhode Island requiring disclosure on request or at a defined point in the hiring process rather than in the posting itself. Several Ohio cities, including Cincinnati, Cleveland, Columbus and Toledo, have local ordinances. Among recent changes, California Senate Bill 642 took effect on 1 January 2026 clarifying the definition of pay scale, Rhode Island added a written notice requirement from 1 January 2026, and Delaware has a requirement scheduled to take effect on 26 September 2027 for employers with 26 or more employees.

The practical technique is to search for your target job title filtered to a covered state, and read the posted ranges even for jobs you would not take and cities you would not move to. You are collecting the employer’s own stated numbers for the level you are targeting. A dozen of those is better evidence than any single survey figure.

Two cautions. Posted ranges are often wide enough to be nearly useless, and the top of a range is frequently reserved for people already in the role rather than new hires. And where a company posts a range, it is under no obligation to offer you anything above the bottom of it.

Putting the sources together

A closing caution

Good data tells you what a job pays. It does not tell you what this employer will pay you, which depends on their budget, their urgency, and whether anyone else is competing for you. A carefully derived number from the 75th percentile is a much better opening position than a guess, and it will still lose to an employer with a fixed band and three other candidates. Knowing the distribution mostly protects you from the opposite error, which is accepting a figure well below the market because nobody told you what the market was.

Where these figures come from

• Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 national estimates and accompanying news release, published 15 May 2026.
• Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2023 national estimates for accountants and auditors, occupation code 13-2011, including metropolitan area estimates.
• Bureau of Labor Statistics, Occupational Employment and Wage Statistics technical note for the May 2025 estimates, and the programme’s published frequently asked questions on the definition of wages and survey scope.
• Jackson Lewis, 2026 overview of pay transparency laws and employer obligations.

General information, current as of September 2026.