Non-Discrimination Testing: Who Fails and Who Pays
Non-discrimination testing decides whether your 401(k) and cafeteria plan keep their tax treatment. The tests, the deadlines, and who pays when one fails.
Non-Discrimination Testing
The annual arithmetic that decides whether your benefit plans keep their tax treatment. The two 401(k) percentage tests, the top-heavy determination, the three cafeteria plan tests, the dependent care rules and the self-insured health pair: who is tested, what passing looks like, what failure costs, and which fixes stop working the moment elections lock
The first time an administrator told me a plan had failed testing, I assumed we had done something wrong. We had not. The documents were right, the deposits on time, the payroll file clean. What had happened was arithmetic.
Three people at the top were saving hard and nobody else was saving at all. Non-discrimination testing is not an audit hunting for mistakes. It is a set of ratios, and when one comes out wrong the tax treatment is taken from the top of the org chart, not the bottom.
Here is the whole map: the two percentage tests, the top-heavy determination, the three cafeteria plan tests, the dependent care rules and the self-insured health pair. I build the people tooling for businesses without an HR department at FirstHR, an onboarding and HR platform rather than a plan administrator. This is general information, not tax advice.
What the Testing Actually Is
Non-discrimination testing is the annual arithmetic that decides whether a benefit plan keeps its tax treatment. Every test compares a favored group against everybody else, and a plan that fails usually loses the tax break for that group rather than its status.
Nobody designs a plan to exclude their own staff. It gets designed generously, the people earning $19 an hour decline to defer 6 percent of it, and the ratios do the rest.
| Test | Which plan | Pass condition | Who pays on failure |
|---|---|---|---|
| Actual deferral percentage test | 401(k) deferrals | Within 1.25 times the other average, or 2 points and no more than double | Highly compensated employees, through taxable refunds |
| Actual contribution percentage test | 401(k) match and after-tax | The same math on match percentages | Same people. Unvested amounts are forfeited |
| Top-heavy determination | The plan as a whole | Key employee balances at 60 percent or less | The employer, through a minimum non-key contribution |
| Coverage test | The plan as a whole | A wide enough cross section of other staff benefits | The employer, who must widen coverage |
| Cafeteria eligibility test | Section 125 plan | Eligibility does not favor the highly compensated | Highly compensated participants lose the exclusion |
| Contributions and benefits test | Section 125 plan | Neither favors that same group | Highly compensated participants |
| Key employee concentration test | Section 125 plan | Key employees get 25 percent or less of the total | Key employees only, which means the owners |
| Eligibility and benefits tests | Dependent care program | Neither favors the highly compensated | Highly compensated employees, on the election |
| 55 percent average benefits test | Dependent care program | Other average is at least 55 percent of theirs | Highly compensated employees |
| Owner concentration rule | Dependent care program | 25 percent or less goes to more than 5 percent owners | The owners |
| Eligibility test | Self-insured health plan | 70 percent of all benefit, or 80 percent of those eligible | Highly compensated individuals, on excess reimbursement |
| Benefits test | Self-insured health plan | Favored benefits are open to all on the same terms | Highly compensated individuals |
The table above maps every test to its plan, its pass condition and the person who pays. It omits one thing: a fully insured group health plan skips section 105(h) entirely, and the parallel rule for insured plans has been unenforced since Notice 2011-1.
Who Counts as Favored
There is no single definition of the favored group. The retirement tests use highly compensated employee, top-heavy uses key employee, and the cafeteria and self-insured health rules each use a further definition matching neither.
The lookback mechanic trips people up more than the thresholds do. For most retirement testing, whether somebody is highly compensated this year turns on what they earned last year, so a person promoted in January is generally not in the group until the following year. Ownership counts the current year too, and every figure resets annually (IRS cost of living adjustments).
The Actual Deferral Percentage Test
The actual deferral percentage test compares the average deferral rate of highly compensated employees against the average for everybody else. It passes if the favored average is no more than 1.25 times the other average, or no more than two percentage points higher and no more than double it (26 CFR 1.401(k)-2).
How the averages get built is what decides small business outcomes. You take each eligible person's deferral rate, then a plain average of those rates. Anyone eligible who defers nothing enters as a zero, and one zero counts as much as one enthusiastic saver.
| Non-highly compensated average | 1.25 times limit | 2 points or 2 times, whichever is less | Ceiling for the favored group |
|---|---|---|---|
| 1.0 percent | 1.25 percent | 2.0 percent (2 times) | 2.00 percent |
| 2.0 percent | 2.50 percent | 4.0 percent (2 times) | 4.00 percent |
| 3.0 percent | 3.75 percent | 5.0 percent (2 points) | 5.00 percent |
| 4.0 percent | 5.00 percent | 6.0 percent (2 points) | 6.00 percent |
| 5.0 percent | 6.25 percent | 7.0 percent (2 points) | 7.00 percent |
| 6.0 percent | 7.50 percent | 8.0 percent (2 points) | 8.00 percent |
Read the last column as your real limit. If the rest of your workforce averages 2 percent, your owners are capped at 4 percent, which on $200,000 of pay is $8,000 against a 2026 deferral limit of $24,500. The statutory limit was never the constraint.
You also pick a method. Prior year testing measures this year favored average against last year other average, so the target is known twelve months ahead. Current year testing compares both in the same year.
The Actual Contribution Percentage Test
The actual contribution percentage test runs identical arithmetic on employer matching contributions and any after-tax employee contributions. Same two groups, same two-part limit, different numerator. A plan can pass one and fail the other.
When the match is a straight formula tied to deferrals, the tests move together and a deferral failure drags the matching test with it. They separate when the match is discretionary, when a year-end true-up applies, or when after-tax contributions exist.
Correction has one wrinkle: only the vested portion of an excess aggregate contribution can be distributed, and anything unvested is forfeited. A plan with no match and no after-tax money skips this test entirely, worth knowing before adding a match to a plan already near the line.
The Top-Heavy Test
A plan is top-heavy when key employee account balances exceed 60 percent of total balances on the determination date, the last day of the preceding plan year. Failing it taxes nobody. It obliges the employer to fund up to 3 percent of pay for every non-key participant, under Internal Revenue Code section 416.
Notice what is measured. It counts accumulated balances rather than current year contributions, so a plan can pass both percentage tests and still be top-heavy because the founder has been saving for eight years.
The minimum has a useful ceiling: it is capped at the highest percentage any key employee received, so if the owners take nothing, nothing is owed. Status is determined annually, so a plan can flip when two long-tenured non-key employees cash out and shrink the denominator.
The Safe Harbor Escape
A safe harbor 401(k) satisfies both percentage tests automatically, and in a plan year holding nothing but deferrals and safe harbor contributions it is also exempt from the top-heavy minimum. The price is an employer contribution that is mandatory and immediately vested.
You are buying certainty with a fixed cost, worth doing when refunds already go back to your owners every March and much less when the plan passes.
The relief people lose without noticing is the top-heavy one, and it is conditional on the year rather than the plan. Allocate discretionary profit sharing on top of the safe harbor money, or forfeitures the same way, and the top-heavy rules return for that year. Which happens in a good year, exactly when an owner wants to.
The Three Cafeteria Plan Tests
A Section 125 cafeteria plan runs three tests: eligibility, contributions and benefits, and key employee concentration. The plan does not collapse when one fails. The favored group loses the exclusion and their elections become taxable wages, while everybody else is unaffected (26 U.S.C. 125).
The concentration test is the one small businesses fail. It asks whether key employees receive more than 25 percent of aggregate qualified benefits. In a company of nine where two owners hold family coverage and four staff waive onto a spouse plan, that ratio arrives on its own.
Timing is specific: the tests run as of the last day of the plan year, counting everyone employed on any day during it. One structural fix exists beforehand. A premium-only plan can use the statutory eligibility safe harbor and skip the other two, so keeping premiums separate from the flexible spending accounts is a testing choice.
The Dependent Care Tests
A dependent care assistance program carries four requirements: an eligibility test, a contributions and benefits test, a 55 percent average benefits test, and a rule that no more than 25 percent of the money may go to more than 5 percent owners (26 U.S.C. 129).
The last two fail for the same structural reason. The people who elect a dependent care account can set aside thousands a year and have young children, which at a small company means the owners. The denominator is small and they are most of it.
The 55 percent test averages the whole eligible population rather than participants, so employees who elect nothing count as zero. Failure turns the excluded amount into taxable wages while the money stays locked in an account that only reimburses care. The 2026 cap is $7,500.
The Self-Insured Health Plan Test
Internal Revenue Code section 105(h) applies to any self-insured medical reimbursement plan, which sweeps in level-funded arrangements, most health reimbursement arrangements, and any employer paying claims from its own funds. It runs eligibility and benefits tests against a favored group defined nowhere else.
Eligibility is satisfied three ways: the plan benefits 70 percent or more of all employees; or 70 percent are eligible and at least 80 percent of those benefit; or it covers a classification the IRS finds nondiscriminatory. You may exclude staff under three years of service, under 25, and part-time or seasonal.
The benefits test asks whether the benefits themselves are the same. A plan that waives the deductible for executives fails here even when everybody is eligible. Failure produces excess reimbursement, taxed as wages to that person alone, which is why it survives quietly for years.
When to Run Each Test
Every test is formally run after the plan year ends, and almost every useful fix happens before it starts. Two dates carry the weight: the close of open enrollment, when the free fixes expire, and the two and a half month mark, when the excise tax starts.
The asymmetry between the two sides is the thing to hold on to. Retirement failures have a correction mechanism with statutory deadlines, so a bad March result is unpleasant but solvable. Cafeteria plan and dependent care failures have no equivalent, because the elections were irrevocable.
The Fixes, and Which Ones Expire
There are two categories of fix and the difference is money. Design fixes are free and only work before elections lock. Corrective fixes work afterwards and cost either taxable income to your owners or real contributions to everybody else.
| Fix | Which tests it helps | Deadline | What it costs |
|---|---|---|---|
| Cap what owners and highly compensated people elect | Every test here | Before open enrollment closes | Nothing in cash. They get less pre-tax room |
| Raise participation through communication | Every test here | Before the plan year starts | Time, plus the higher take-up you asked for |
| Add automatic enrollment to the 401(k) | Both percentage tests | Amendment before the year, or a year ahead under prior year testing | Higher match spend as participation climbs |
| Switch to a safe harbor design | Percentage tests, top-heavy in a clean year | Generally before the plan year starts | A mandatory, immediately vested contribution |
| Qualified nonelective contribution | Both percentage tests | 12 months after year end, under current year testing | Employer money, vested, to people who did not ask |
| Corrective refunds to the favored group | Both percentage tests | 2.5 months after year end, 6 with automatic enrollment | Taxable income for the people you meant to help |
| Top-heavy minimum contribution | Top-heavy determination | By the employer filing deadline with extensions | Up to 3 percent of pay for non-key participants |
| Restructure eligibility or split the plan | Coverage and eligibility tests | Before the plan year starts | Amendments, and a harder conversation about who gets what |
The first row is the one nobody wants and almost always the cheapest. Telling two owners to elect $4,000 into a dependent care account instead of $7,500 costs the business nothing. Discovering in February that the whole election is taxable costs them more, and costs you the conversation anyway.
Frequently Asked Questions
What is non-discrimination testing?
Non-discrimination testing is the annual set of calculations that decides whether a benefit plan keeps its favorable tax treatment. Each test compares what a favored group receives against what everyone else receives, and the favored group is defined differently depending on the test. Congress attached these conditions to the tax breaks: employers get pre-tax treatment for retirement deferrals, health premiums and dependent care provided the arrangement does not exist mainly for the people who wrote it. Failure rarely touches the rank and file. It removes the tax break from owners, officers and highly compensated employees, or it obliges the employer to fund a contribution for everybody else.
Which benefit plans have to be tested?
A 401(k) runs two percentage tests on deferrals and matching contributions, a coverage test, and an annual top-heavy determination. A Section 125 cafeteria plan runs an eligibility test, a contributions and benefits test, and a key employee concentration test. A dependent care assistance program adds an eligibility test, a benefits test, a 55 percent average benefits test and an owner concentration rule. A self-insured medical reimbursement plan, which covers most level-funded arrangements and health reimbursement arrangements, runs eligibility and benefits tests under Internal Revenue Code section 105(h). Group term life insurance and educational assistance carry their own rules. A fully insured group health plan is the main thing that escapes all of it.
What happens if a 401(k) fails the deferral test?
The excess has to come back out. The standard correction refunds the excess contributions and attributable earnings to the highly compensated employees who created them, and those refunds are taxable to them in the year received. If the refunds happen more than two and a half months after the plan year ends, the employer owes a 10 percent excise tax on the excess under Internal Revenue Code section 4979, and a plan with an eligible automatic contribution arrangement gets six months instead. The alternative is a qualified nonelective contribution to the non-highly compensated group, which is real employer money, immediately vested, and available up to twelve months after year end.
Who is a highly compensated employee for 401(k) testing?
Two routes get you into the group. The first is ownership: anyone who owned more than 5 percent of the business at any point in the current or preceding plan year is highly compensated regardless of pay, and family attribution rules can pull in a spouse, child or parent. The second is pay above an indexed threshold in the lookback year, which IRS Notice 2025-67 sets at $160,000 for 2026. An employer may also make a top-paid group election, which limits the compensation route to the highest paid 20 percent of the workforce and helps a business with several well-paid people and no owners among them.
What makes a 401(k) plan top-heavy?
A plan is top-heavy when key employee account balances exceed 60 percent of total plan balances on the determination date, which is the last day of the preceding plan year. The test measures accumulated balances rather than current year contributions, which is why young plans at small companies are so often top-heavy: the owner has been saving longer or rolled a prior balance in, and the denominator is small. Failing it taxes nobody. It obliges the employer to make a minimum contribution of up to 3 percent of pay to every non-key participant, capped at the highest contribution percentage any key employee received.
Do small businesses have to run cafeteria plan testing?
Yes, and the concentration test is the one that catches them. A Section 125 plan fails when key employees receive more than 25 percent of the aggregate qualified benefits, and at a company where two owners hold family coverage while half the staff waives, that ratio arrives without anyone doing anything unusual. The consequence lands on the key employees alone: their elections become taxable wages while everyone else keeps the exclusion. A premium-only plan can use the statutory eligibility safe harbor and skip the other two tests, which is a practical reason to keep the premium arrangement separate from the flexible spending accounts.
When should non-discrimination testing be run?
Formally after the plan year ends, practically before it starts. The tests measure a completed year, so the official run happens in the first quarter for a calendar year plan, and the retirement corrections have hard deadlines at two and a half months and twelve months after year end. Every cheap fix lives on the other side of the calendar. Capping what owners elect, adding automatic enrollment and pushing participation all have to happen before open enrollment closes, because cafeteria plan elections become irrevocable at that point. Model the tests in late summer on projected numbers, then run them for real in the new year.