How to Find Real Estate Comps Without Guessing
Pulling twenty candidate properties takes four minutes. Deciding which four are real is the entire job — and it's where most pricing quietly goes wrong.
Short answer
Screen in this order: arm's-length → recency → boundaries → property type → size band → condition. Then apply the substitution test — would a buyer have genuinely considered this instead of the subject? — and disqualify anything needing more than about 15% in net adjustments. Twenty candidates should come down to three to six. If it doesn't, your screen is too loose.
If you're looking for where to pull comps from — MLS, public records, aggregators, automated reports — that's how to find real estate comps: 7 methods that actually work. This piece assumes you have candidates already and covers the harder half: deciding which ones survive.
Why the screen matters more than the source
Two agents pull from the same MLS on the same morning and produce prices $40,000 apart. The data was identical. The screening wasn't.
That gap is almost never caused by a missing source. It's caused by one agent keeping a comp that shouldn't have survived — an estate sale, a house across the school boundary, something with a finished basement counted as living area. One bad comp in a set of four moves the answer by a quarter of the total error.
Adding a fifth mediocre comp does not make a comp set more reliable. It makes it wider, and a wider range is easier to argue with.
The screen, in order
Order matters. Each step is cheaper than the one after it, so run them in sequence and stop wasting attention on candidates already eliminated.
Step 1 — Arm's length, or out
First and non-negotiable. A sale only tells you about market value if both parties were acting in their own interest with normal exposure to the market.
Out immediately:
- Family transfers and sales between related parties
- Estate and probate sales (often priced for speed, not value)
- Foreclosures, REO and short sales
- Sales between an owner and their own trust or LLC
- Anything with a price that looks like a round-number settlement rather than a negotiation
The MLS won't always flag these. A $310,000 sale in a $420,000 street is telling you something — go find out what before you use it or discard it.
Step 2 — Recency
Three to six months in a normal market. Stretch to twelve only when the set is genuinely thin, and apply a market-conditions adjustment if prices moved in between.
| Age of sale | Use it? | Caveat |
|---|---|---|
| 0–3 months | Ideal | None |
| 3–6 months | Good | Check whether the market moved |
| 6–12 months | Only if thin | Market-conditions adjustment required |
| 12 months+ | Rarely | Support only, never a primary comp |
Remember that a closed sale reflects a price agreed 30–60 days before closing. A "three-month-old" comp is really a four-and-a-half-month-old negotiation. In a moving market that gap matters, which is why pendings belong in the report alongside solds.
Step 3 — Boundaries, not radius
The most common screening error, and the one automated tools make most reliably.
Distance is a proxy for similarity, and it's a bad one. A house 0.3 miles away across a school district line is not more comparable than one 1.1 miles away inside it. Things that break comparability regardless of distance:
- School attendance boundaries
- Municipal or jurisdiction lines (different taxes, services, sometimes different rules)
- Major arterials, rail lines, highways — physical severances
- HOA vs. non-HOA, and different HOAs with different dues
- Flood zone designations
- Subdivision identity, where buyers perceive one as distinct
Half a mile is a reasonable starting radius in a suburban tract and a meaningless one in rural or mixed-density areas. Start with the boundary map, not the circle.
Step 4 — Property type and configuration
Match the things buyers won't trade off: detached vs. attached, single-story vs. two-story where the market cares, bedroom count within one, and total bathroom count within one.
Watch the single-story premium specifically — in markets with older demographics it can be substantial and is easy to under-adjust.
Step 5 — Size band
Keep above-grade square footage within roughly 10–15% of the subject. Beyond that you're relying on square-footage adjustments to do more work than they reliably can, because price per square foot is non-linear — a 3,000 sq ft home does not sell for twice a 1,500 sq ft home.
Two verification points that catch a lot of errors:
- Below-grade space isn't above-grade space. A finished basement adds value at a fraction of the rate. If the record lumps them together, unpick it.
- Tax record and MLS often disagree. When they do, find out which is right before using either. This is a frequent source of quiet error.
Step 6 — Condition and updates
The step that can't be done from a desk, and the reason automated comp selection stays a starting point rather than an answer.
Look at the listing photos of every surviving comp. A comp that sold with a 1980s kitchen is not comparable to your subject's 2023 renovation, whatever the specs say. If the photos are gone, treat the comp as lower confidence.
The substitution test
After the mechanical screen, apply the one judgment test that catches what rules miss:
If this comp hadn't existed, would its buyer have plausibly bought the subject property instead?
That's the definition of comparable, and it catches things no filter does. A house that's technically similar but sits on a busy road, backs a commercial lot, or has a 40-foot power easement across the yard fails substitution even when it passes every numeric screen.
Disqualifiers after the screen
Some comps survive screening and should still be dropped:
| Signal | Why it disqualifies |
|---|---|
| Net adjustments over ~15% | The property is telling you it isn't comparable |
| Heavy seller concessions | Recorded price overstates what the market paid |
| Unverifiable square footage | Every adjustment built on it inherits the error |
| Extreme days on market | Sold under duress or mispriced throughout |
| Sold off-market with no exposure | Never tested against the market |
| Only comp supporting your number | Not a rule — but check yourself hard here |
That last row is the honest one. If dropping a single comp collapses your price, you don't have a price — you have a preference with one data point under it.
What "without guessing" actually means
Guessing isn't picking the wrong comp. It's picking a comp for a reason you couldn't say out loud to the seller.
The test for every comp that makes your report: can you finish the sentence "I included this one because…" with something factual? "Same subdivision, same floor plan, sold nine weeks ago, similar condition" is a reason. "It was in the list and the number looked about right" is a guess wearing a suit.
Which is also why the adjustments have to be visible in the report itself — showing the arithmetic is what proves the selection wasn't arbitrary. See CMA adjustments explained.
When there genuinely aren't enough comps
Sometimes the honest answer is that the data is thin. Rural properties, acreage, heavy custom builds, unusual configurations, brand-new subdivisions.
Do this rather than padding:
- Widen time before geography. An older sale of a genuinely similar house beats a recent sale of a different one.
- Widen to comparable submarkets with a location adjustment you can defend, not a nearby but dissimilar area.
- Use actives and expireds to bracket. When solds are thin, the ceiling and the competition still carry information.
- Say so. A report stating "three comps, thin market, wider range accordingly" is more credible than one presenting six comps of which three are fiction.
The bottom line
Finding comps is a search problem and it's basically solved — the sources are known and mostly a click away. Screening them is a judgment problem, and judgment is what you're actually paid for.
Run the six steps in order, apply substitution, drop anything over 15% net adjustment, and be able to say out loud why each survivor is in. Do that and the price defends itself, because the reasoning is visible rather than asserted.
Frequently asked questions
What makes a property a good comp?
A good comp is a recent arm's-length closed sale of a similar property that a buyer would have genuinely considered as an alternative to the subject. The practical test is substitution: if this house had not sold, would that buyer have plausibly bought the subject instead?
How recent do comps need to be?
Three to six months in a normal market. Push to twelve months only when the comp set is genuinely thin, and apply a market-conditions adjustment if prices moved. In a fast-moving market, a six-month-old sale can be materially stale.
How close do comps need to be?
Distance matters less than boundaries. Half a mile is a common guideline in a suburban tract, but a comp two blocks away across a school district line, a major arterial or a jurisdiction boundary is often worse than one a mile away inside the same boundaries.
What disqualifies a comparable sale?
Non-arm's-length transactions — family transfers, estate sales, foreclosures, short sales, and any sale where the price reflects the circumstances rather than the market. Also unverifiable square footage, undisclosed heavy concessions, and any property needing more than roughly 15% in net adjustments.
How many comps do you need?
Three to six closed sales. Fewer than three gives no pattern to read. More than six usually means the screen was too loose and weak comparables were added to pad the count, which widens the range rather than narrowing it.