How to Choose the Right Sample Size for Your Survey

Updated on April 23, 2026

Every researcher, marketer, or student who has ever designed a survey hits the same wall: how many people do I actually need to survey? Too few and your results are meaningless noise. Too many and you've burned budget and time for no real gain in accuracy. Getting this right is less about guesswork and more about three concrete inputs — confidence level, margin of error, and population size — that feed into a formula you can apply today.

Let's walk through it properly, with real numbers.

Why Sample Size Actually Matters

Picture this: you run a small business and want to know if customers prefer your new packaging over the old one. You ask five people. Three say yes — that's 60% approval. But if you'd asked a sixth person and they said no, you'd suddenly be at 50%. That swing of nearly 10 percentage points from a single response is exactly the kind of instability that makes small samples dangerous for real decisions.

Flip side: a national polling organization doesn't need to survey all 330 million Americans to get reliable results. They typically use around 1,000–1,500 respondents and achieve a margin of error of roughly ±3%. Sample size doesn't scale linearly with population — and that's one of the most counterintuitive and useful facts in statistics.

The Three Inputs You Need

1. Confidence Level

Your confidence level is the probability that your survey results actually capture the true value in the population. The most common choices are 90%, 95%, and 99%.

In practice, 95% is the standard for most research. It means that if you ran your survey 100 times with different random samples, 95 of those results would fall within your stated margin of error. The remaining 5 might not — and you'd never know which run you're on.

Each confidence level maps to a Z-score:

  • 90% confidence → Z = 1.645
  • 95% confidence → Z = 1.96
  • 99% confidence → Z = 2.576

Higher confidence means a larger required sample. That trade-off is real and worth thinking about for your specific situation. For an internal employee satisfaction survey, 90% might be fine. For a peer-reviewed study or a high-stakes product launch decision, you'll want 95% or higher.

2. Margin of Error

Margin of error (sometimes called the confidence interval) is how much variance you're willing to tolerate in either direction. A margin of error of ±5% means if your survey shows 60% approval, the true number is probably somewhere between 55% and 65%.

Common values:

  • ±3% — tight, professional-grade surveys (used in political polling)
  • ±5% — the workhorse; acceptable for most business and academic surveys
  • ±10% — rough estimates, exploratory research, or when resources are very limited

Cutting your margin of error in half — say from 10% to 5% — quadruples your required sample size. That's not a typo. The relationship is inverse and squared. So before automatically reaching for ±3%, ask yourself: does my decision actually require that precision, or would ±5% produce the same action?

3. Population Size

Here's where people often overcomplicate things. Population size matters much less than you'd expect, especially once your population exceeds about 20,000 people.

For very large or effectively infinite populations (say, all adults in a city, or all users of a website with millions of visitors), population size has almost no effect on your required sample. The formula essentially ignores it.

For small, finite populations — 500 employees at a company, 2,000 customers of a niche product — a finite population correction actually lets you survey fewer people than you would otherwise need.

The Formula

For an infinite or very large population, the core formula is:

n = (Z² × p × (1 − p)) / e²

Where:

  • n = required sample size
  • Z = Z-score for your confidence level
  • p = estimated proportion (use 0.5 if unknown — this maximizes the sample size and keeps you conservative)
  • e = margin of error as a decimal (5% = 0.05)

Let's run through a real example.

Worked Example: Customer Satisfaction Survey

You work at a SaaS company. You want to survey your customers about satisfaction with a new feature. You want 95% confidence, you can live with ±5% margin of error, and you don't know what the satisfaction rate will be (so you'll use p = 0.5).

Plugging in:

  • Z = 1.96
  • p = 0.5
  • e = 0.05

n = (1.96² × 0.5 × 0.5) / 0.05²
n = (3.8416 × 0.25) / 0.0025
n = 0.9604 / 0.0025
n = 384.16 → round up to 385

So you need at least 385 completed responses. Always round up — never down.

Now suppose your company only has 1,200 customers total. You apply the finite population correction:

n_adjusted = n / (1 + (n − 1) / N)
n_adjusted = 385 / (1 + 384 / 1200)
n_adjusted = 385 / (1 + 0.32)
n_adjusted = 385 / 1.32
n_adjusted = 291.7 → round up to 292

Because your population is finite and relatively small, you actually only need 292 responses instead of 385. That's a meaningful reduction in outreach effort.

The Part Most Guides Skip: Response Rate

Here's the practical catch that breaks a lot of otherwise well-planned surveys: the number above is your target for completed responses. But most people don't respond to every survey they receive.

If you expect a 30% response rate and you need 385 completes, you'll need to send your survey to:

385 / 0.30 = 1,283 people

Response rates vary wildly depending on context. Internal employee surveys often hit 40–70% with a well-timed nudge from leadership. Cold email surveys to customers might see 10–20%. Adjust your outreach list accordingly, or you'll collect far fewer responses than you planned for.

Practical Decision Guide

Here's how to think through your choices before touching any formula:

  1. What decision hinges on this data? High-stakes decisions (product pivots, significant budget allocation, public reporting) warrant higher confidence and tighter margins. Internal pulse checks can tolerate more flexibility.
  2. Do you have any prior data? If a previous survey showed roughly 70% satisfaction, use p = 0.7 in your formula. This typically reduces your required n compared to the conservative 0.5 assumption.
  3. What's your actual population size? If it's under 10,000, run the finite correction. If it's over 100,000, it barely matters — use the base formula.
  4. What's realistic to collect? Working backwards from budget and time constraints to see if your targets are feasible is not cheating — it's good design. Sometimes you discover you need to either tighten resources or widen your margin of error.

Using a Sample Size Calculator

You don't need to run this arithmetic by hand every time. A good online sample size calculator handles all of this instantly — you input confidence level, margin of error, and population size, and it spits out the required n. The real value of understanding the formula isn't to do mental math; it's so you know why the calculator returns what it does, and so you can sanity-check outputs that look wrong.

For example, if a calculator tells you that you need 16,576 responses for a 99% confidence level with ±1% margin of error, you'll understand that's not an error — it's the brutal cost of that extreme precision. You can then make an informed decision to loosen one of the inputs.

A Few Things That Won't Fix a Bad Sample

No amount of statistical rigor saves a survey from a biased sampling method. If you email your survey only to customers who rated you 5 stars last month, your results reflect that group — not your full customer base. Sample size and sample quality are separate problems. A well-chosen sample of 300 beats a convenience sample of 3,000 almost every time.

Also: survey fatigue is real. A 40-question survey sent to 500 people will likely get worse data than a 10-question survey sent to the same list. Completion rates drop, and the people who finish long surveys may not be representative of the broader group.

Putting It Together

To land on a defensible sample size, start by anchoring your decision in the three core inputs: pick a confidence level that matches the stakes of your work, choose a margin of error that reflects how much uncertainty you can actually tolerate, and account for your population size if it's on the smaller side. Run the formula (or use a calculator), then divide your target completions by your expected response rate to get your outreach number.

The goal isn't a massive sample — it's a right-sized one. A 385-person survey with careful random sampling will often tell you more than a 5,000-person survey full of selection bias. Get the math right, get the method right, and your results will actually hold up when someone asks how you got there.